Greg Thompson
Add initial intent classification model to nlu endpoint
0dc3012
raw
history blame
7.56 kB
from fuzzywuzzy import fuzz
from mathtext_fastapi.logging import prepare_message_data_for_logging
from mathtext.sentiment import sentiment
from mathtext.text2int import text2int
from mathtext_fastapi.intent_classification import create_intent_classification_model, retrieve_intent_classification_model, predict_message_intent
import re
def build_nlu_response_object(type, data, confidence):
""" Turns nlu results into an object to send back to Turn.io
Inputs
- type: str - the type of nlu run (integer or sentiment-analysis)
- data: str/int - the student message
- confidence: - the nlu confidence score (sentiment) or '' (integer)
>>> build_nlu_response_object('integer', 8, 0)
{'type': 'integer', 'data': 8, 'confidence': 0}
>>> build_nlu_response_object('sentiment', 'POSITIVE', 0.99)
{'type': 'sentiment', 'data': 'POSITIVE', 'confidence': 0.99}
"""
return {'type': type, 'data': data, 'confidence': confidence}
# def test_for_float_or_int(message_data, message_text):
# nlu_response = {}
# if type(message_text) == int or type(message_text) == float:
# nlu_response = build_nlu_response_object('integer', message_text, '')
# prepare_message_data_for_logging(message_data, nlu_response)
# return nlu_response
def test_for_number_sequence(message_text_arr, message_data, message_text):
""" Determines if the student's message is a sequence of numbers
>>> test_for_number_sequence(['1','2','3'], {"author_id": "57787919091", "author_type": "OWNER", "contact_uuid": "df78gsdf78df", "message_body": "I am tired", "message_direction": "inbound", "message_id": "dfgha789789ag9ga", "message_inserted_at": "2023-01-10T02:37:28.487319Z", "message_updated_at": "2023-01-10T02:37:28.487319Z"}, '1, 2, 3')
{'type': 'integer', 'data': '1,2,3', 'confidence': 0}
>>> test_for_number_sequence(['a','b','c'], {"author_id": "57787919091", "author_type": "OWNER", "contact_uuid": "df78gsdf78df", "message_body": "I am tired", "message_direction": "inbound", "message_id": "dfgha789789ag9ga", "message_inserted_at": "2023-01-10T02:37:28.487319Z", "message_updated_at": "2023-01-10T02:37:28.487319Z"}, 'a, b, c')
{}
"""
nlu_response = {}
if all(ele.isdigit() for ele in message_text_arr):
nlu_response = build_nlu_response_object(
'integer',
','.join(message_text_arr),
0
)
prepare_message_data_for_logging(message_data, nlu_response)
return nlu_response
def run_text2int_on_each_list_item(message_text_arr):
""" Attempts to convert each list item to an integer
Input
- message_text_arr: list - a set of text extracted from the student message
Output
- student_response_arr: list - a set of integers (32202 for error code)
>>> run_text2int_on_each_list_item(['1','2','3'])
[1, 2, 3]
"""
student_response_arr = []
for student_response in message_text_arr:
int_api_resp = text2int(student_response.lower())
student_response_arr.append(int_api_resp)
return student_response_arr
def run_sentiment_analysis(message_text):
""" Evaluates the sentiment of a student message
>>> run_sentiment_analysis("I am tired")
[{'label': 'NEGATIVE', 'score': 0.9997807145118713}]
>>> run_sentiment_analysis("I am full of joy")
[{'label': 'POSITIVE', 'score': 0.999882698059082}]
"""
# TODO: Add intent labelling here
# TODO: Add logic to determine whether intent labeling or sentiment analysis is more appropriate (probably default to intent labeling)
return sentiment(message_text)
def run_intent_classification(message_text):
""" Process a student's message using basic fuzzy text comparison
>>> run_intent_classification("exit")
{'type': 'intent', 'data': 'exit', 'confidence': 1.0}
>>> run_intent_classification("exi")
{'type': 'intent', 'data': 'exit', 'confidence': 0.86}
>>> run_intent_classification("eas")
{'type': 'intent', 'data': '', 'confidence': 0}
>>> run_intent_classification("hard")
{'type': 'intent', 'data': '', 'confidence': 0}
>>> run_intent_classification("hardier")
{'type': 'intent', 'data': 'harder', 'confidence': 0.92}
"""
label = ''
ratio = 0
nlu_response = {'type': 'intent', 'data': label, 'confidence': ratio}
commands = [
'easier',
'exit',
'harder',
'hint',
'next',
'stop',
]
for command in commands:
ratio = fuzz.ratio(command, message_text.lower())
if ratio > 80:
nlu_response['data'] = command
nlu_response['confidence'] = ratio / 100
return nlu_response
def evaluate_message_with_nlu(message_data):
""" Process a student's message using NLU functions and send the result
>>> evaluate_message_with_nlu({"author_id": "57787919091", "author_type": "OWNER", "contact_uuid": "df78gsdf78df", "message_body": "8", "message_direction": "inbound", "message_id": "dfgha789789ag9ga", "message_inserted_at": "2023-01-10T02:37:28.487319Z", "message_updated_at": "2023-01-10T02:37:28.487319Z"})
{'type': 'integer', 'data': 8, 'confidence': 0}
>>> evaluate_message_with_nlu({"author_id": "57787919091", "author_type": "OWNER", "contact_uuid": "df78gsdf78df", "message_body": "I am tired", "message_direction": "inbound", "message_id": "dfgha789789ag9ga", "message_inserted_at": "2023-01-10T02:37:28.487319Z", "message_updated_at": "2023-01-10T02:37:28.487319Z"})
{'type': 'sentiment', 'data': 'NEGATIVE', 'confidence': 0.9997807145118713}
"""
# Keeps system working with two different inputs - full and filtered @event object
try:
message_text = message_data['message_body']
except KeyError:
message_data = {
'author_id': message_data['message']['_vnd']['v1']['chat']['owner'],
'author_type': message_data['message']['_vnd']['v1']['author']['type'],
'contact_uuid': message_data['message']['_vnd']['v1']['chat']['contact_uuid'],
'message_body': message_data['message']['text']['body'],
'message_direction': message_data['message']['_vnd']['v1']['direction'],
'message_id': message_data['message']['id'],
'message_inserted_at': message_data['message']['_vnd']['v1']['chat']['inserted_at'],
'message_updated_at': message_data['message']['_vnd']['v1']['chat']['updated_at'],
}
message_text = message_data['message_body']
# Run intent classification only for keywords
intent_api_response = run_intent_classification(message_text)
if intent_api_response['data']:
return intent_api_response
number_api_resp = text2int(message_text.lower())
if number_api_resp == 32202:
# Run intent classification with logistic regression model
predicted_label = predict_message_intent(message_text)
if predicted_label['confidence'] > 0.01:
nlu_response = predicted_label
return nlu_response
# Run sentiment analysis
sentiment_api_resp = sentiment(message_text)
nlu_response = build_nlu_response_object(
'sentiment',
sentiment_api_resp[0]['label'],
sentiment_api_resp[0]['score']
)
else:
nlu_response = build_nlu_response_object(
'integer',
number_api_resp,
0
)
prepare_message_data_for_logging(message_data, nlu_response)
return nlu_response