def get_zylon_parser_scores(self):
"""
parameters: none
Extracts labelled entities from zylon's xml output and true xml
output. Compares the entity lists and returns a score, higher is
better.
return: edu_insts_match_score, edu_majors_match_score, emp_names_match_score, emp_jtitles_match_score
"""
extractor = Extractor()
zylon_filenames = extractor.populate_file_names(
self.__zylon_parser_labels_folder)
zylon_xml_trees = extractor.read_resume_labels(
self.__zylon_parser_labels_folder, zylon_filenames)
true_xml_trees = extractor.read_resume_labels(
self.__dataset_raw_folder, zylon_filenames)
true_edu_insts = [
extractor.get_edu_institutions(xml_tree)
for xml_tree in true_xml_trees
]
true_edu_majors = [
extractor.get_edu_majors(xml_tree) for xml_tree in true_xml_trees
]
true_emp_names = [
extractor.get_company_names(xml_tree)
for xml_tree in true_xml_trees
]
true_emp_jtitles = [
extractor.get_job_titles(xml_tree) for xml_tree in true_xml_trees
]
zylon_edu_insts = [
extractor.get_edu_institutions_zy(xml_tree)
for xml_tree in zylon_xml_trees
]
zylon_edu_majors = [
extractor.get_edu_majors_zy(xml_tree)
for xml_tree in zylon_xml_trees
]
zylon_emp_names = [
extractor.get_company_names_zy(xml_tree)
for xml_tree in zylon_xml_trees
]
zylon_emp_jtitles = [
extractor.get_job_titles_zy(xml_tree)
for xml_tree in zylon_xml_trees
]
tokeniser = Tokeniser()
true_edu_insts = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(true_edu_insts))
true_edu_majors = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(true_edu_majors))
true_emp_names = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(true_emp_names))
true_emp_jtitles = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(true_emp_jtitles))
zylon_edu_insts = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(zylon_edu_insts))
zylon_edu_majors = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(zylon_edu_majors))
zylon_emp_names = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(zylon_emp_names))
zylon_emp_jtitles = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(zylon_emp_jtitles))
edu_insts_match_score = self.score_matches(zylon_edu_insts,
true_edu_insts)
edu_majors_match_score = self.score_matches(zylon_edu_majors,
true_edu_majors)
emp_names_match_score = self.score_matches(zylon_emp_names,
true_emp_names)
emp_jtitles_match_score = self.score_matches(zylon_emp_jtitles,
true_emp_jtitles)
return edu_insts_match_score, edu_majors_match_score, emp_names_match_score, emp_jtitles_match_score
def get_ies_scores(self):
extractor = Extractor()
ies_filenames = extractor.populate_file_names(self.__ies_accuracy_test)
ies_filenames = extractor.filter_by_valid_exts(ies_filenames)
filenames, resume_content = extractor.read_resume_content_tika_api(
ies_filenames, self.__ies_accuracy_test)
filenames, resume_content = extractor.remove_empty_resumes(
filenames, resume_content)
resume_labels = extractor.read_resume_labels(self.__ies_accuracy_test,
filenames)
true_edu_insts = [
extractor.get_edu_institutions(xml_tree)
for xml_tree in resume_labels
]
true_edu_majors = [
extractor.get_edu_majors(xml_tree) for xml_tree in resume_labels
]
true_emp_names = [
extractor.get_company_names(xml_tree) for xml_tree in resume_labels
]
true_emp_jtitles = [
extractor.get_job_titles(xml_tree) for xml_tree in resume_labels
]
cs = CrfSuite()
cs.load_tagger()
annotator = Annotator()
annotated_resumes = [
annotator.annotate_using_trained_model(self.__ies_accuracy_test +
self.__seperator +
filename[0] + filename[1])
for filename in filenames
]
predicted_entity_list = [
cs.tag_doc(resume) for resume in annotated_resumes
]
ies_edu_insts = [
extractor.get_edu_institutions_from_list(entity_list)
for entity_list in predicted_entity_list
]
ies_edu_majors = [
extractor.get_edu_major_from_list(entity_list)
for entity_list in predicted_entity_list
]
ies_emp_names = [
extractor.get_company_names_from_list(entity_list)
for entity_list in predicted_entity_list
]
ies_emp_jtitles = [
extractor.get_company_position_from_list(entity_list)
for entity_list in predicted_entity_list
]
tokeniser = Tokeniser()
true_edu_insts = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(true_edu_insts))
true_edu_majors = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(true_edu_majors))
true_emp_names = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(true_emp_names))
true_emp_jtitles = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(true_emp_jtitles))
ies_edu_insts = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(ies_edu_insts))
ies_edu_majors = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(ies_edu_majors))
ies_emp_names = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(ies_emp_names))
ies_emp_jtitles = tokeniser.docs_tolower(
tokeniser.tokenise_doclines_to_words(ies_emp_jtitles))
edu_insts_match_score = self.score_matches(ies_edu_insts,
true_edu_insts)
edu_majors_match_score = self.score_matches(ies_edu_majors,
true_edu_majors)
emp_names_match_score = self.score_matches(ies_emp_names,
true_emp_names)
emp_jtitles_match_score = self.score_matches(ies_emp_jtitles,
true_emp_jtitles)
print(edu_insts_match_score)
print(edu_majors_match_score)
print(emp_names_match_score)
print(emp_jtitles_match_score)
class Annotator():
__job_position_tag = "EMP-POS"
__job_company_tag = "EMP-COMP"
__education_course_tag = "EDU-MAJOR"
__education_institution_tag = "EDU-INST"
def __init__(self):
self.__extractor = Extractor()
self.__tokeniser = Tokeniser()
self.__tagger = Tagger()
self.__dataset = Dataset()
self.__logger = Logger()
def prepare_dataset(self, nr_of_docs=-1):
resumes, labels = self.__extractor.read_raw_files(nr_of_docs)
resumes = self.__tokeniser.tokenise_docs_to_lines(resumes)
resumes = self.__tokeniser.tokenise_doclines_to_words(resumes)
self.__dataset.resume_content = self.annotate_docs(resumes, labels)
self.__dataset.save()
# resumes: list of tokenised (by line and word) résumé docs
# labels: xml structure storing labels for several resumes
def annotate_docs(self, resumes, labels):
self.__logger.println("annotating resumes")
annotated_resumes = []
for idx, resume in enumerate(resumes):
annotated_resumes.append(self.annotate_doc(resume, labels[idx]))
self.__logger.println(
"annotating resume %s/%s with true labels and pos tags" %
(idx + 1, len(resumes)))
# non local ner tag entire dataset at a time for speed
annotated_resumes = self.__tagger.nonlocal_ner_tag(annotated_resumes)
self.__logger.println("completed annotating resumes")
return annotated_resumes
# doc: a single résumé document with token strings in each slot of list
# labels: xml structure storing pre-extracted information
def annotate_doc(self, doc, labels):
job_title_list = self.__extractor.get_job_titles(labels)
job_company_list = self.__extractor.get_company_names(labels)
edu_major_list = self.__extractor.get_edu_majors(labels)
edu_inst_list = self.__extractor.get_edu_institutions(labels)
# can extract more labels here
prepared_doc = self.__tagger.prepare_doc(doc)
prepared_doc = self.__match_entity(prepared_doc, job_title_list,
self.__job_position_tag)
prepared_doc = self.__match_entity(prepared_doc, job_company_list,
self.__job_company_tag)
prepared_doc = self.__match_entity(prepared_doc, edu_major_list,
self.__education_course_tag)
prepared_doc = self.__match_entity(prepared_doc, edu_inst_list,
self.__education_institution_tag)
prepared_doc = self.__tagger.add_default_entity_tags(prepared_doc)
prepared_doc = self.__tagger.pos_tag(prepared_doc)
return prepared_doc
# doc: résumé doc to be annotated
# entity_list: list of labels to matched in doc
# tag: tag to be assigned if match found
def __match_entity(self, doc, entity_list, tag):
for entity in entity_list:
doc = self.__tagger.match_label(doc, entity, tag)
return doc
# function takes in a path to file and annotates it for tagging
# to be ideally used to tag as a one off for testing
# filepath: path to résumé
def annotate_using_trained_model(self, filepath):
resume_content = self.__extractor.read_resume_content(filepath)
resume_content = self.__tokeniser.tokenise_docs_to_lines(
resume_content)
resume_content = self.__tokeniser.tokenise_doclines_to_words(
resume_content)
prepared_doc = self.__tagger.prepare_doc(resume_content[0])
prepared_doc = self.__tagger.pos_tag(prepared_doc)
prepared_doc = self.__tagger.nonlocal_ner_tag([prepared_doc])
return prepared_doc[0]
