Spam Classification using OpenAI
The majority of people in today’s society own a mobile phone, and they all frequently get communications (SMS/email) on their phones. But the key point is that some of the messages you get may be spam, with very few being genuine or important interactions. You may be tricked into providing your personal information, such as your password, account number, or Social Security number, by scammers that send out phony text messages. They may be able to access your bank, email, and other accounts if they obtain this information. To filter out these messages, a spam filtering system is used that marks a message spam on the basis of its contents or sender.
In this article, we will be seeing how to develop a spam classification system and also evaluate our model using various metrics. In this article, we will be majorly focusing on OpenAI API. There are 2 ways to
We will be using the Email Spam Classification Dataset dataset which has mainly 2 columns and 5572 rows with spam and non-spam messages.
Steps to implement Spam Classification using OpenAI
Now there are two approaches that we will be covering in this article:
1. Using Embeddings API developed by OpenAI
Step 1: Install all the necessary salaries
!pip install -q openai
Step 2: Import all the required libraries
Python3
# necessary libraries import openai import pandas as pd import numpy as np # libraries to develop and evaluate a machine learning model from sklearn.ensemble import RandomForestClassifier from sklearn.model_selection import train_test_split from sklearn.metrics import classification_report, accuracy_score from sklearn.ensemble import RandomForestClassifier from sklearn.model_selection import train_test_split from sklearn.metrics import classification_report, accuracy_score from sklearn.metrics import confusion_matrix |
Step 3: Assign your API key to the OpenAI environment
Python3
# replace "YOUR API KEY" with your generated API key openai.api_key = "YOUR API KEY" |
Step 4: Read the CSV file and clean the dataset
Our dataset has 3 unnamed columns with NULL values,
Note: Open AI’s public API does not process more than 60 requests per minute. so we will drop them and we are taking only 60 records here only.
Python3
# while loading the csv, we ignore any encoding errors and skip any bad line df = pd.read_csv( 'spam.csv' , encoding_errors = 'ignore' , on_bad_lines = 'skip' ) print (df.shape) # we have 3 columns with NULL values, to remove that we use the below line df = df.dropna(axis = 1 ) # we are taking only the first 60 rows for developing the model df = df.iloc[: 60 ] # rename the columns v1 and v2 to Output and Text respectively df.rename(columns = { 'v1' : 'OUTPUT' , 'v2' : 'TEXT' }, inplace = True ) print (df.shape) df.head() |
Output:
Step 5: Define a function to use Open AI’s Embedding API
We use the Open AI’s Embedding function to generate embedding vectors and use them for classification. Our API uses the “text-embedding-ada-002” model which belongs to the second generation of embedding models developed by OpenAI. The embeddings generated by this model are of length 1536.
Python3
# function to generate vector for a string def get_embedding(text, model = "text-embedding-ada-002" ): return openai.Embedding.create( input = , model = model)[ 'data' ][ 0 ][ 'embedding' ] # applying the above funtion to generate vectors for all 60 text pieces df[ "embedding" ] = df.TEXT. apply (get_embedding). apply (np.array) # convert string to array df.head() |
Output:
Step 6: Custom Label the classes of the output variable to 1 and 0, where 1 means “spam” and 0 means “not spam”.
Python3
class_dict = { 'spam' : 1 , 'ham' : 0 } df[ 'class_embeddings' ] = df.OUTPUT. map (class_dict) df.head() |
Output:
Step 7: Develop a Classification model.
We will be splitting the dataset into a training set and validation dataset using train_test_split and training a Random Forest Classification model.
Python3
# split data into train and test X = np.array(df.embedding) y = np.array(df.class_embeddings) X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2 , random_state = 42 ) # train random forest classifier clf = RandomForestClassifier(n_estimators = 100 ) clf.fit(X_train.tolist(), y_train) preds = clf.predict(X_test.tolist()) # generate a classification report involving f1-score, recall, precision and accuracy report = classification_report(y_test, preds) print (report) |
Output:
precision recall f1-score support
0 0.82 1.00 0.90 9
1 1.00 0.33 0.50 3
accuracy 0.83 12
macro avg 0.91 0.67 0.70 12
weighted avg 0.86 0.83 0.80 12
Step 8: Calculate the accuracy of the model
Python3
print ( "accuracy: " , np. round (accuracy_score(y_test, preds) * 100 , 2 ), "%" ) |
Output:
accuracy: 83.33 %
Step 9: Print the confusion matrix for our classification model
Python3
confusion_matrix(y_test, preds) |
Output:
array([[9, 0],
[2, 1]])
2. Using text completion API developed by OpenAI
!pip install -q openai
Step 2: Import the following libraries
Python3
import openai |
Step 3: Assign your API key to the Openaithe environment
Python3
# replace "YOUR API KEY" with your generated API key openai.api_key = "YOUR API KEY" |
Step 4: Define a function using the text completion API of Openai
Python3
def spam_classification(message): response = openai.Completion.create( model = "text-davinci-003" , prompt = f "Classify the following message as spam or not spam:\n\n{message}\n\nAnswer:" , temperature = 0 , max_tokens = 64 , top_p = 1.0 , frequency_penalty = 0.0 , presence_penalty = 0.0 ) return response[ 'choices' ][ 0 ][ 'text' ].strip() |
Step 5: Try out the function with some examples
Example 1:
Python3
out = spam_classification( """Congratulations! You've Won a $1000 gift card from walmart. Go to https://bit.ly to claim your reward.""" ) print (out) |
Output:
Spam
Example 2:
Python3
out = spam_classification( "Hey Alex, just wanted to let you know tomorrow is an off. Thank you" ) print (out) |
Output:
Not spam
Frequently Asked Questions (FAQs)
1. Which algorithm is best for spam detection?
There isn’t a single algorithm that has consistently produced reliable outcomes. The type of the spam, the data that is accessible, and the particular requirements of the problem are some of the variables that affect an algorithm’s efficiency. Although Naive Bayes, Neural Networks (RNNs), Logistic Regression, Random Forest, and Support Vector Machines are some of the most frequently used classification techniques.
2. What is embedding or word embedding?
The embedding or Word embedding is a natural language processing (NLP) technique where words are mapped into vectors of real numbers. It is a way of representing words and documents through a dense vector representation. This representation is learned from data and is shown to capture the semantic and syntactic properties of words. The words closest in vector space have the most similar meanings.
3. Is spam classification supervised or unsupervised?
Spam classification is supervised as one requires both independent variable(message contents) and target variables(outcome,i.e., whether the email is spam or not) to develop a model.
4. What is spam vs ham classification?
Email that is not spam is referred to be “Ham”. Alternatively, “good mail” or “non-spam” It ought to be viewed as a quicker, snappier alternative to “non-spam”. The phrase “non-spam” is probably preferable in most contexts because it is more extensively used by anti-spam software makers than it is elsewhere.
Conclusion
In this article, we discussed the development of a spam classifier using OpenAI modules. Open AI has many such modules that can help you ease your daily work and also help you get started with projects in the field of Artificial Intelligence. You can check out other tutorials using Open AI API’s below: