feat(answer_chooser): implement basic chooser with AnswerChooser

knowledge_verificator/answer_chooser.py

@@ -0,0 +1,85 @@
+"""Module with AnswerChooser, which finds a best candidate for an answer in a paragraph."""
+
+import random
+import nltk
+from nltk.corpus import wordnet
+
+# FIXME: Write docstrings.
+
+
+class AnswerChooser:
+    def __init__(self) -> None:
+        nltk.download('wordnet')
+        nltk.download('stopwords')
+        nltk.download('punkt_tab')
+
+    def remove_stopwords(self, text: str) -> str:
+        """
+        Remove stopwords from a string using NLTK.
+        """
+        # Get the stopwords from the NLTK stopwords corpus
+        stopwords = set(nltk.corpus.stopwords.words('english'))
+
+        # Tokenize the text into words
+        words = nltk.word_tokenize(text)
+
+        # Remove the stopwords from the words
+        filtered_words = [
+            word for word in words if word.lower() not in stopwords
+        ]
+
+        # Join the filtered words back into a string
+        cleaned_text = ' '.join(filtered_words)
+
+        # Return the cleaned text
+        return cleaned_text
+
+    def santize(self, word: str) -> str:
+        """Convert to lowercase and remove any punctuation mark."""
+        word = word.strip()
+        word = word.lower()
+        to_remove = ('.', ',', '?', '!', '-', '_', '/', '(', ')', "'")
+        for punctuation_mark in to_remove:
+            word = word.replace(punctuation_mark, '')
+        return word
+
+    def find_part_of_speech(self, word: str) -> str:
+        """
+        Determine the part of speech of a word using WordNet.
+        """
+        # Look up the word in WordNet
+        word = self.santize(word=word)
+        synsets = wordnet.synsets(word)
+
+        # If the word is not found, return 'n/a'
+        if not synsets:
+            return 'n/a'
+
+        # Get the first synset and determine the part of speech
+        synset = synsets[0]
+        pos = synset.pos()
+
+        # Map WordNet POS tags to more common POS tags
+        if pos == 'a':
+            return 'adjective'
+        elif pos == 'n':
+            return 'noun'
+        elif pos == 'r':
+            return 'adverb'
+        elif pos == 'v':
+            return 'verb'
+        else:
+            return 'n/a'
+
+    def choose_answer(self, paragraph: str) -> str:
+        paragraph = self.remove_stopwords(paragraph)
+        words = paragraph.split(' ')
+        # FIXME: Refactor not to use three times the same function...
+        words = [
+            self.santize(word)
+            for word in words
+            if self.santize(word)
+            and self.find_part_of_speech(self.santize(word)) == 'n/a'
+        ]
+
+        return random.choice(words)

knowledge_verificator/main.py

@@ -1,30 +1,32 @@
 """Main module with CLI definition."""
 
-import random
 import logging
 
 from logging import Logger
 
+from knowledge_verificator.answer_chooser import AnswerChooser
 from knowledge_verificator.nli import NaturalLanguageInference, Relation
 from knowledge_verificator.qg import QuestionGeneration
 
 
 if __name__ == '__main__':
-    # TODO: Take logger level from CLI or config file.
     logger = Logger('main_logger')
+    # TODO: Take logger level from CLI or config file.
     # Set logging to standard output (handle 2.) stream.
     logging_handler = logging.StreamHandler()
     logging_handler.setLevel(logging.DEBUG)
     logger.addHandler(logging_handler)
 
+    chooser = AnswerChooser()
+    qg_module = QuestionGeneration()
+
     paragraph = input('Enter a paragraph you would like to learn: ')
     logger.debug('Loaded the following paragraph:\n %s', paragraph)
 
-    qg_module = QuestionGeneration()
-
     # Answer is a randomly choosen word.
-    words = paragraph.split(' ')
-    chosen_answer = random.choice(words)
+    chosen_answer = chooser.choose_answer(paragraph=paragraph)
+    # words = paragraph.split(' ')
+    # chosen_answer = random.choice(words)
     logger.debug(
         'The `%s` has been chosen as the answer, based on which the question will be generated.',
         chosen_answer,
@@ -40,7 +42,9 @@
         'Question Generation module has supplied the question: %s', question
     )
 
-    user_answer = input(f'Answer the question. {question}')
+    user_answer = input(
+        f'Answer the question with full sentence. {question} \n Your answer.: '
+    )
 
     nli_module = NaturalLanguageInference()
     relation = nli_module.infer_relation(
@@ -53,6 +57,6 @@
         case Relation.CONTRADICTION:
             feedback = f'wrong. Correct answer is {chosen_answer}'
         case Relation.NEUTRAL:
-            feedback = 'not directly answer the posed question.'
+            feedback = 'not directly associated with the posed question.'
 
     print(f'Your answer is {feedback}')

pyproject.toml

@@ -11,6 +11,8 @@ transformers = "^4.44.2"
 torch = "^2.4.1"
 pylint-pytest = "^1.1.8"
 sentencepiece = "^0.2.0"
+nltk = "^3.9.1"
+rich = "^13.8.1"
 
 [tool.poetry.group.test]