def test_is_spam_function_two_returns_false_vns_test_seven(self): return_val = EmailAnalyzer.is_spam_function_two(False, 80, 80) self.assertFalse(return_val)
def test_is_spam_function_two_returns_true_vns_test_four(self): return_val = EmailAnalyzer.is_spam_function_two(False, 20, 65) self.assertTrue(return_val)
def test_is_spam_function_two_returns_false_pic_test_one(self): return_val = EmailAnalyzer.is_spam_function_two(False, 80, 65) self.assertFalse(return_val)
def test_is_spam_function_two_returns_true_vns_test_three(self): return_val = EmailAnalyzer.is_spam_function_two(True, 20, 80) self.assertTrue(return_val)
class RENEGE: """Class pour realiser le filtrage du spam en utilisant vocabular.json file et CRUD et EmalAnalyze classes""" def __init__(self): self.email_file = "800-mails.json" self.crud = CRUD() self.e_mail = EmailAnalyzer() def calculate_user_trust(self, user_id): #extracting json data date_of_first_seen_message = self.crud.get_user_data(user_id, "Date_of_first_seen_message") date_of_last_seen_message = self.crud.get_user_data(user_id, "Date_of_last_seen_message") n_ham = self.crud.get_user_data(user_id, "HamN") n_spam = self.crud.get_user_data(user_id, "SpamN") groups = self.crud.get_user_data(user_id, "Groups") #calculate the sum of trust values of all groups sum_trust = 0 for group in groups : group_id = self.crud.get_group_id(group) sum_trust += self.crud.get_group_data(group_id, 'Trust') #now that we have all the needed vars, calculate trust1, trust2 and trust trust1 = (date_of_last_seen_message * n_ham) / (date_of_first_seen_message * (n_ham + n_spam)) trust2 = sum_trust / len(groups) trust = (trust1 + trust2) / 2 if trust2 < 50: trust = trust2 if trust1 > 100: trust = 100 #before returning a value, check if trust is between 0 and 100 if trust < 0: trust = 0 elif trust > 100: trust = 100 return trust def classify_emails(self, calculation_mode, is_log_est, is_log_combo): ''' fonction deja implemente Description: fonction pour commencer l'analyse des e-mails. Sortie: bool, 'True' pour succes, 'False' dans le cas de failure. ''' try: self.process_email(self.get_email(), calculation_mode, is_log_est, is_log_combo) return True except Exception: raise Exception def process_email(self, new_emails, calculation_mode, is_log_est, is_log_combo): ''' Description: fonction pour analyser chaque nouvel e-mail dans le dictionare. Elle gere l'ajout des nouveux utilisateurs et/ou modification de l'information existante sur les utilisateurs et groupes. Sortie: bool, 'True' pour succes, 'False' dans le cas de failure. ''' print(type(new_emails)) emails = new_emails["dataset"] for email in emails: email_adr = email['mail']['From'] date = email['mail']['Date'] spam = email['mail']['Spam'] == 'true' subject = email['mail']['Subject'] body = email['mail']['Body'] user_id = self.crud.get_user_id(email_adr) if user_id: self.update_user_info(email_adr, date, spam) else: self.add_user_info(email_adr, date) # trust de l'utilisateur trust = self.crud.get_user_data(user_id, "Trust") # moyenne du trust de tous les groupes user_group = self.crud.get_user_data(user_id, "Groups") sum_trust = 0 groups = self.crud.read_groups_file() for group in groups : if group['List_of_members'] == user_group: sum_trust += group['Trust'] avg_group_trust = sum_trust / len(groups) # difference de jours entre last_seen_msg et first_seen_msg user_activity = self.substract_dates(self.crud.get_user_data(email_adr, "Date_of_last_seen_message"), self.crud.get_user_data(email_adr, "Date_of_first_seen_message")) if(calculation_mode == 1): spam = self.e_mail.is_spam_function_one(spam, user_activity, trust, avg_group_trust) elif(calculation_mode == 2): spam = self.e_mail.is_spam_function_two(spam, trust, avg_group_trust) elif(calculation_mode == 0): spam = self.e_mail.is_spam(subject, body, is_log_est, is_log_combo, 0) # mettre a jour l'utilisateur avec la nouvelle valeur de spam self.update_user_info(email_adr, date, spam) return True def substract_dates(self, last_seen_msg, first_seen_msg): # calcul de difference de jours entre le dernier et le premier message last_seen_msg = str(last_seen_msg).split("-") first_seen_msg = str(first_seen_msg).split("-") user_activity = date(int(last_seen_msg[0]), int(last_seen_msg[1]), int(last_seen_msg[2])) - date(int(first_seen_msg[0]), int(first_seen_msg[1]), int(first_seen_msg[2])) # convertir la soustraction en int user_activity = int((str(user_activity).split(" "))[0]) return user_activity