CI Status Updates Documentation Status


Given a gazetteer/taxonomy and input text, taxonomy-matcher can be used to find all phrases which matches the codes/instances/keywords in the gazetteer or taxonomy.

For each match, it will return the information of,

  • surface_form
  • matched position
  • Code ID and Code Description
  • and other code related information


Python 3.6+


Use taxonomy-match script:

usage: taxonomy-match input_file taxonomy_file [--output_file OUTPUT_FILE]

load taxonomy phrases from the taxonomy file, and find all matched phrases
from the input text. The result will eithor write to an output file or print
to the screen.

positional arguments:
  input_file            input text file, text to mine phrases
  taxonomy_file         taxonomy file, support json/xml/txt, see documentation
                        for more details

optional arguments:
  --output_file         output file of matched phrases, supports
                        jsonl/csv/tsv/txt format, print matched phrases to
                        the screen if not defined

Use taxonomy-matcher module

  • From normalization table in JSOM format:
from taxonomy_matcher.matcher import Matcher
taxonomy_matcher = Matcher(normtable=json_file)
for matched in taxonomy_matcher.matching(text):

And an example of the normalization table in JSON:

  "meta": {
    "concept_type": "skills",
    "release_datetime": "2019-xx-xx"
  "concepts": [
      "display_name": "Risk Analysis",
      "category": "Financial Skill",
      "id": "ABCDEFG001",
      "surface_forms": [
          "surface_form": "risk analysis",
          "skill_likelihood": 0.9
          "surface_form": "quantitative risk assessment",
          "skill_likelihood": 1.0
          "surface_form": "risk assessment",
          "skill_likelihood": 0.7
      "display_name": "Mobile Data",
      "category": "Computer Skill",
      "id": "ABCDEFG002",
      "surface_forms": [
          "surface_form": "mobile data"
  • From gazetteer:
from taxonomy_matcher.matcher import Matcher
taxonomy_matcher = Matcher(gazetteer=gz_file)
for matched in taxonomy_matcher.matching(text):

and an example of the gazetteer

# gazetteer
mobile data
risk analysis
quantitative risk assessment
risk assessment
  • From Taxonomy Codetable:
from taxonomy_matcher.matcher import Matcher
ct_matcher = Matcher(codetable=ct_file)
for matched in ct_matcher.matching(text):

CodeTable is a XML version of the JSON example given above.

other functions

  • Context words:

When context are needed for matched phrases, e.g. for the following up validation functions, enable the with\_context option:

from taxonomy_matcher.matcher import Matcher
taxonomy_matcher = Matcher(normtable=json_file,with_context=True)
for matched in taxonomy_matcher.matching(text):
    print(matched.left_context, matched.right_context)
  • Code Property lookup

If need to lookup the property of an Code in the taxonomy, check the matcher Class property ‘code_property_mapping’, it is a dictionary mapping id to description and category, it is in the form of:

dict[code_id] = {

E.g. to get the description of the codeid:

codeid = 12345
from taxonomy_matcher.matcher import Matcher
taxonomy_matcher = Matcher(normtable=json_file)
if codeid in taxonomy_matcher.code_property_mapping:

check the Metainfo of the Taxonomy or Gazetteer:

Note: currently only available for the Normalized code JSOM.

The metainfo can be stored in meta part of the JSON document, e.g. if the following information is listed in the JSOM meta section:

"meta": {
  "language": "EN",
  "release_datetime": "2019-04-17T12:22:10.729673",
  "concept_type": "skills",
  "purpose": "normalization"

We can fetch it via the matcher object

from taxonomy_matcher.matcher import Matcher
taxonomy_matcher = Matcher(normtable=json_file)

output will be:

  'language': 'EN',
  'release_datetime': '2019-04-17T12:22:10.729673',
  'concept_type': 'skills',
  'purpose': 'normalization'

matched phrase object: MatchedPhrase

matcher.matching is an iterable which return a MatchedPhrase instance, the instance has the following attributes:

  • normalize pattern form: matched_pattern
  • surface form: surface_form
  • start position and end position: start_pos, end_pos
  • code_id and code_description (None if not set in the pattern file)
  • left context and right context of the matched skills (only availabe if with_context=True )
for match in matcher.matching(text):
    print("found pattern [{}] in the form of [{}] at position ({}:{}), code:{} {} {}".format(


To install package and its dependencies, run the following from project root directory:

python install


To run unit tests, execute the following from the project root directory:

python test