From c3b642e5bb2b7fc0d0326c95ba01817e2562366c Mon Sep 17 00:00:00 2001
From: amau345 <amaury.tiravy@ecl20.ec-lyon.fr>
Date: Wed, 31 Jan 2024 17:13:59 +0100
Subject: [PATCH] new_similarity_tests
---
test_combine.py | 342 +++++++++++++++++++++++++++
test_similarity.py => test_jacard.py | 29 ++-
testwordembeding.py | 33 ---
3 files changed, 358 insertions(+), 46 deletions(-)
create mode 100644 test_combine.py
rename test_similarity.py => test_jacard.py (94%)
delete mode 100644 testwordembeding.py
diff --git a/test_combine.py b/test_combine.py
new file mode 100644
index 0000000..1a829df
--- /dev/null
+++ b/test_combine.py
@@ -0,0 +1,342 @@
+def calculate_combined_score(tfidf_score, jaccard_score):
+ # You can adjust the weights based on the importance of each score
+ return 0.7 * tfidf_score + 0.3 * jaccard_score
+
+def get_best_answers(question, text_lines, vectorizer, vectorial_base):
+
+ question_vector = vectorizer.transform([question]).toarray()
+
+ # Calculate cosine similarity between the question and each text line
+ similarities = cosine_similarity(question_vector, vectorial_base).flatten()
+
+ jaccard_similarities = [jaccard_similarity(question, text) for text in text_lines]
+
+ # Calculate TF-IDF score for each text line
+ tfidf_scores = [vectorizer.transform([text]).toarray() for text in text_lines]
+
+ # Calculate combined scores using both TF-IDF and cosine similarity
+ combined_scores = [calculate_combined_score(tfidf_score, jaccard_score)
+ for tfidf_score, jaccard_score in zip(tfidf_scores, jaccard_similarities)]
+
+ # Get the indices of the top 3 most similar text lines based on the combined scores
+ top_indices = np.argsort(combined_scores)[-3:][::-1]
+
+ # Retrieve the corresponding text lines along with their combined scores
+ best_answers = [text_lines[i]+"\n" for i in top_indices]
+
+ return best_answers
+
+import sys
+from PyQt5.QtWidgets import QApplication, QWidget, QHBoxLayout, QVBoxLayout, QTextEdit, QLineEdit, QPushButton, QSizePolicy, QListWidget, QListWidgetItem, QLabel
+from PyQt5.QtCore import Qt
+from PyQt5.QtGui import QPalette, QColor, QFont, QIcon
+from sklearn.feature_extraction.text import TfidfVectorizer
+from sklearn.metrics.pairwise import cosine_similarity
+import numpy as np
+from nltk.corpus import stopwords
+from nltk.tokenize import word_tokenize
+
+def read_text_file(file_path):
+ """
+ Reads the content of a text file specified by `file_path` and splits it into paragraphs based on double line breaks (`'\n\n'`).
+ Parameters:
+ - file_path (str): The path to the text file.
+ Returns:
+ - list: A list of non-empty paragraphs from the file.
+ """
+ with open(file_path, 'r', encoding='utf-8') as file:
+ content = file.read().split('\n\n')
+ content1 = [item for item in content if item != ""]
+ return content1
+
+def extract_keywords_french(sentence):
+ """
+ Tokenizes and filters a given sentence to extract keywords in French. Removes stop words and focuses on meaningful terms.
+ Parameters:
+ - sentence (str): The input sentence.
+ Returns:
+ - str: A string containing the extracted keywords.
+ """
+ stop_words = set(stopwords.words('french'))
+ mots_questions = ['qui', 'quoi', 'où', 'quand', 'pourquoi', 'comment', 'quel', 'quelle', 'quels', 'quelles', 'est-ce que', 'y a-t-il', 'peut-on', 'sont-ils', 'sont-elles', 'combien', 'lequel', 'laquelle', 'lesquels', 'lesquelles', 'est-ce', 'n\'est-ce pas', 'savoir', 'pouvez-vous', 'êtes-vous', 'avez-vous', 'dois-je', 'quelqu\'un', 'quelque chose']
+ stop_words=stop_words.union(mots_questions)
+ words = word_tokenize(sentence, language='french')
+ keywords = [word for word in words if word.lower() not in stop_words]
+ return ' '.join(keywords)
+
+
+def create_vectorial_base(text_lines, min_chars=10):
+ """
+ Creates a TF-IDF vectorial base from a list of text lines.
+ Parameters:
+ - text_lines (list): List of text lines.
+ - min_chars (int): Minimum number of characters required for a line to be included (default is 10).
+ Returns:
+ - tuple: A tuple containing the TF-IDF vectorizer, the TF-IDF matrix (vectorial base), and the feature names.
+ """
+ filtered_lines = [line for line in text_lines if len(line) >= min_chars]
+
+ if not filtered_lines:
+ print("No lines with at least 10 characters found.")
+ return None, None, None
+
+ vectorizer = TfidfVectorizer() #a tester en option : stop_words=list(stopwords.words('french'))
+ vectorial_base = vectorizer.fit_transform(filtered_lines).toarray()
+ feature_names = vectorizer.get_feature_names_out()
+
+ return vectorizer, vectorial_base, feature_names
+
+def jaccard_similarity(str1, str2):
+ tokens_str1 = set(word_tokenize(str1.lower()))
+ tokens_str2 = set(word_tokenize(str2.lower()))
+
+ stop_words = set(stopwords.words('english'))
+ tokens_str1 = tokens_str1 - stop_words
+ tokens_str2 = tokens_str2 - stop_words
+
+ intersection = len(tokens_str1.intersection(tokens_str2))
+ union = len(tokens_str1) + len(tokens_str2) - intersection
+
+ similarity = intersection / union if union != 0 else 0.0
+ return similarity
+
+def get_best_answers(question, text_lines, vectorizer, vectorial_base):
+ """
+ Retrieves the top 3 most similar text lines to a given question based on cosine similarity.
+ Parameters:
+ - question (str): The user's question.
+ - text_lines (list): List of text lines.
+ - vectorizer: The TF-IDF vectorizer.
+ - vectorial_base: The TF-IDF matrix (vectorial base).
+ Returns:
+ - list: A list of the top 3 most similar text lines as answers.
+ """
+ question_vector = vectorizer.transform([question]).toarray()
+
+ # Calculate cosine similarity between the question and each text line
+ similarities = cosine_similarity(question_vector, vectorial_base).flatten()
+
+ jaccard_similarities = [jaccard_similarity(question, text) for text in text_lines]
+
+ combined_scores = [calculate_combined_score(similarities, jaccard_score)
+ for similarities, jaccard_score in zip(similarities, jaccard_similarities)]
+
+ # Get the indices of the top 3 most similar text lines
+ top_indices = np.argsort(combined_scores)[-3:][::-1]
+ # Retrieve the corresponding text lines
+ best_answers = [text_lines[i]+"\n" for i in top_indices]
+
+ return best_answers
+
+class WrappingLabel(QLabel):
+ """
+ Subclass of QLabel with word wrapping enabled. Used for displaying text in the GUI.
+ """
+ def __init__(self, text='', parent=None):
+ super(WrappingLabel, self).__init__(text, parent)
+ self.setWordWrap(True)
+
+
+class StyledListWidgetItem(QListWidgetItem):
+ """
+ Subclass of QListWidgetItem with custom styling for the chat history list.
+ """
+ def __init__(self, text='', parent=None):
+ super(StyledListWidgetItem, self).__init__(parent)
+ self.setText(text)
+
+ def initStyle(self):
+ palette = QPalette()
+ palette.setColor(QPalette.Highlight, QColor("#4b5261")) # Couleur de fond pour l'élément sélectionné dans la liste d'historique
+ palette.setColor(QPalette.HighlightedText, QColor("#ff0000")) # Couleur du texte pour l'élément sélectionné dans la liste d'historique
+ self.setData(Qt.UserRole, palette)
+
+class StyledListWidget(QListWidget):
+ """
+ Subclass of QListWidget with custom styling for the chat history list.
+ """
+ def __init__(self, parent=None):
+ super(StyledListWidget, self).__init__(parent)
+ self.setAlternatingRowColors(False)
+ self.setStyleSheet("""
+ QListWidget {
+ background-color: #282c34; /* Couleur de fond pour la liste d'historique */
+ color: #abb2bf; /* Couleur du texte dans la liste d'historique */
+ border-radius: 10px; /* Coins arrondis */
+ }
+ """)
+
+ def addStyledItem(self, text):
+ """
+ Adds a styled item to the list widget.
+ Parameters:
+ - text (str): The text to be added to the list.
+ """
+ item = StyledListWidgetItem(text)
+ item.initStyle()
+ self.addItem(item)
+
+
+class ChatbotInterface(QWidget):
+ """
+ Main class representing the chatbot interface. Initializes the UI and handles user interactions.
+ """
+ def __init__(self):
+ super().__init__()
+ file_path = 'reglementdescolarite-ingegeneraliste2324-1.docx.txt'
+ self.text_lines = read_text_file(file_path)
+
+ if not self.text_lines:
+ print("The file is empty or doesn't exist.")
+ return
+ self.vectorizer, self.vectorial_base, _ = create_vectorial_base(self.text_lines)
+ self.init_ui()
+ self.command_history = [] # Pour stocker l'historique des commandes
+ self.dico={}
+ self.dico2={}
+
+ def init_ui(self):
+ """
+ Initializes the user interface.
+ """
+ # Créer des widgets
+ self.conversation_text = QTextEdit(self)
+ self.conversation_text.setFont(QFont("consolas",9))
+ self.conversation_text.setReadOnly(True)
+
+ self.user_input_entry = QLineEdit(self)
+ self.user_input_entry.setPlaceholderText("Saisissez votre message...")
+ self.user_input_entry.setMinimumHeight(40)
+
+ self.send_button = QPushButton("Envoyer", self)
+ self.send_button.setMinimumSize(self.user_input_entry.width(), 30) # Ajustez selon vos besoins
+ self.send_button.setMaximumSize(200, 60)
+ self.send_button.clicked.connect(self.send_message)
+
+ # Historique à droite
+ self.history_list_widget = StyledListWidget(self)
+ self.history_list_widget.itemClicked.connect(self.history_item_clicked)
+ self.history_list_widget.setFixedWidth(200) # Ajuster la largeur selon vos besoins
+
+ # Configurer la mise en page
+ layout = QVBoxLayout(self)
+ h_layout = QHBoxLayout()
+
+ # Widgets à gauche
+ left_layout = QVBoxLayout()
+ left_layout.addWidget(self.conversation_text)
+ left_layout.addWidget(self.user_input_entry)
+
+ # Ajouter le bouton "Envoyer" avec une taille réduite
+ self.send_button.setMaximumWidth(self.send_button.width() // 3)
+ left_layout.addWidget(self.send_button, alignment=Qt.AlignRight)
+ h_layout.addLayout(left_layout)
+
+ # Historique à droite
+ h_layout.addWidget(self.history_list_widget)
+ layout.addLayout(h_layout)
+
+ # Configurer la politique de taille pour permettre à la zone de conversation de s'étendre verticalement
+ size_policy = QSizePolicy(QSizePolicy.Preferred, QSizePolicy.Expanding)
+ self.conversation_text.setSizePolicy(size_policy)
+
+ # Définir la fenêtre principale
+ icon = QIcon("chatbot.png")
+ self.setWindowIcon(icon)
+ self.setWindowTitle('chatbot')
+ self.setGeometry(100, 100, 800, 600)
+
+ # Appliquer les styles
+ self.setStyleSheet("""
+ QWidget {
+ background-color: #282c34; /* Couleur principale de fond pour l'application */
+ color: #abb2bf; /* Couleur du texte principal */
+ }
+
+ QTextEdit, QLineEdit {
+ background-color: #2c313a; /* Couleur de fond pour la zone de texte et d'entrée utilisateur */
+ color: #abb2bf; /* Couleur du texte dans la zone de texte et d'entrée utilisateur */
+ border-radius: 10px; /* Coins arrondis */
+ }
+
+ QPushButton {
+ background-color: #61afef; /* Couleur de fond pour le bouton Envoyer */
+ color: #282c34; /* Couleur du texte sur le bouton Envoyer */
+ border-radius: 10px; /* Coins arrondis */
+ }
+
+
+ """)
+
+ self.user_input_entry.returnPressed.connect(self.send_message)
+ self.history_list_widget.itemClicked.connect(self.history_item_clicked)
+
+ def send_message(self):
+ """
+ Handles the user's input, processes it, and displays the chatbot's response.
+ """
+ user_command = self.user_input_entry.text()
+ if len(user_command)>0:
+ self.conversation_text.clear()
+ self.conversation_text.append(f"demande élève: {user_command}")
+ self.conversation_text.append("Réponse du chatbot pour la demande: ")
+
+ best_answers = get_best_answers(user_command, self.text_lines, self.vectorizer, self.vectorial_base)
+ chatbot_response=""
+ for i, answer in enumerate(best_answers, start=1):
+ chatbot_response+=(f"{i}. {answer.strip()}\n\n")
+ self.conversation_text.append(chatbot_response)
+
+ # Ajouter la commande à l'historique
+ user_command1=extract_keywords_french(user_command)
+ self.command_history.append(user_command1)
+ self.dico2[user_command1]= user_command
+ self.dico[user_command1]= chatbot_response
+
+ # Mettre à jour la liste d'historique
+ self.update_history_list()
+
+ self.user_input_entry.clear()
+ else:
+ pass
+
+ def update_history_list(self):
+ """
+ Updates the chat history list in the UI.
+ """
+ self.history_list_widget.clear()
+ for command in self.command_history:
+ self.history_list_widget.addStyledItem(command)
+
+
+ def history_item_clicked(self, item):
+ """
+ Displays the chat history when an item is clicked.
+ Parameters:
+ - item: The clicked item.
+ """
+ self.conversation_text.clear()
+ # Réafficher le contenu dans la conversation_text lorsque l'élément de l'historique est cliqué
+ selected_index = self.history_list_widget.row(item)
+ if selected_index < len(self.command_history):
+ selected_command = self.command_history[selected_index]
+ self.conversation_text.append(f"demande élève: {self.dico2[selected_command]}")
+ # Traiter la commande et obtenir la réponse du chatbot (vous devrez ajuster cela en fonction de votre application)
+ chatbot_response = f"Réponse du chatbot pour la demande: \n{self.dico[selected_command]}"
+ self.conversation_text.append(chatbot_response)
+
+if __name__ == '__main__':
+ app = QApplication(sys.argv)
+ chatbot_app = ChatbotInterface()
+ screen = app.primaryScreen()
+
+ # Ajuster la taille de la fenêtre
+ new_width = screen.availableGeometry().width()*3 // 5
+ chatbot_app.resize(new_width, int(screen.availableGeometry().height()-48))
+
+ # Centrer la fenêtre
+ center_point = screen.availableGeometry().center().x()-chatbot_app.rect().center().x()
+ chatbot_app.move(center_point,0)
+
+ chatbot_app.show()
+ sys.exit(app.exec_())
\ No newline at end of file
diff --git a/test_similarity.py b/test_jacard.py
similarity index 94%
rename from test_similarity.py
rename to test_jacard.py
index 0b23267..18d2fda 100644
--- a/test_similarity.py
+++ b/test_jacard.py
@@ -58,25 +58,28 @@ def create_vectorial_base(text_lines, min_chars=10):
return vectorizer, vectorial_base, feature_names
+def calculate_combined_score(tfidf_score, similarity_score):
+ # You can adjust the weights based on the importance of each score
+ return 0.7 * tfidf_score + 0.3 * similarity_score
+
def get_best_answers(question, text_lines, vectorizer, vectorial_base):
- """
- Retrieves the top 3 most similar text lines to a given question based on cosine similarity.
- Parameters:
- - question (str): The user's question.
- - text_lines (list): List of text lines.
- - vectorizer: The TF-IDF vectorizer.
- - vectorial_base: The TF-IDF matrix (vectorial base).
- Returns:
- - list: A list of the top 3 most similar text lines as answers.
- """
+
question_vector = vectorizer.transform([question]).toarray()
# Calculate cosine similarity between the question and each text line
similarities = cosine_similarity(question_vector, vectorial_base).flatten()
- # Get the indices of the top 3 most similar text lines
- top_indices = np.argsort(similarities)[-3:][::-1]
- # Retrieve the corresponding text lines
+ # Calculate TF-IDF score for each text line
+ tfidf_scores = [vectorizer.transform([text]).toarray() for text in text_lines]
+
+ # Calculate combined scores using both TF-IDF and cosine similarity
+ combined_scores = [calculate_combined_score(tfidf_score, similarity)
+ for tfidf_score, similarity in zip(tfidf_scores, similarities)]
+
+ # Get the indices of the top 3 most similar text lines based on the combined scores
+ top_indices = np.argsort(combined_scores)[-3:][::-1]
+
+ # Retrieve the corresponding text lines along with their combined scores
best_answers = [text_lines[i]+"\n" for i in top_indices]
return best_answers
diff --git a/testwordembeding.py b/testwordembeding.py
deleted file mode 100644
index 717a6d6..0000000
--- a/testwordembeding.py
+++ /dev/null
@@ -1,33 +0,0 @@
-import torch
-from transformers import CamembertForQuestionAnswering, CamembertTokenizer
-
-def answer_question(question, context):
- # Load pre-trained CamemBERT model and tokenizer
- model_name = 'camembert-base' # You can choose a different model if needed
- tokenizer = CamembertTokenizer.from_pretrained(model_name)
- model = CamembertForQuestionAnswering.from_pretrained(model_name)
-
- # Tokenize input question and context
- inputs = tokenizer(question, context, return_tensors='pt', max_length=512, truncation=True)
-
- # Perform question answering
- outputs = model(**inputs)
- start_scores = outputs.start_logits
- end_scores = outputs.end_logits
-
- # Get the answer span
- answer_start = torch.argmax(start_scores)
- answer_end = torch.argmax(end_scores) + 1
- answer = tokenizer.convert_tokens_to_string(tokenizer.convert_ids_to_tokens(inputs['input_ids'][0][answer_start:answer_end]))
-
- return answer
-
-if __name__ == '__main__':
- # Example usage
- user_question = "qui compose le jury ?"
- with open('reglementdescolarite-ingegeneraliste2324-1.docx.txt', 'r', encoding='utf-8') as file:
- content = file.read()
- passage = content
-
- result = answer_question(user_question, passage)
- print(f"Réponse : {result}")
--
GitLab