Text Classification

Text Classifier learns the characteristics of each label from the training data to assign labels to new data.

They generally feature:

1. Input

   • document, $d$
   • predefined set of classes $\{c_1,c_2,c_3,...,c_j\}$

2. Output

   • predefined class $c\in C$

Here are the ML models for text classification:

• Naive Bayes (NB)
• Support Vector Machine (SVM)
• Extreme Learning Machines (ELM)
• Guassian Processes (GP)
• Logistic Regression

Classification predicts a category. Regression predicts a numerical target, such as a review score. Logistic regression is a classifier despite its name.

Classification needs labels. When there are no labels, documents are grouped by similarity instead, which is what K-Means Clustering, Hierarchical Clustering (HC) and Fuzzy Clustering do. Once a classifier has produced an output, its quality is measured with Evaluation Metrics.

---

Code

From text to predicted labels

from sklearn.pipeline import make_pipeline
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.naive_bayes import MultinomialNB
 
texts = ["good fun film", "good story", "bad dull film", "bad story"]
labels = [1, 1, 0, 0]
classifier = make_pipeline(CountVectorizer(), MultinomialNB())
classifier.fit(texts, labels)
print(classifier.predict(["good film", "bad film"]).tolist())  # [1, 0]

Read the pipeline:

• texts contains four training documents; labels contains their sentiment classes
• CountVectorizer learns the vocabulary and converts documents into count rows
• MultinomialNB learns from those rows and their labels
• fit(texts, labels) performs training
• predict applies the same fitted vocabulary to the new documents before classification
• predict also uese the same pipeline under make_pipeline as training
• Each new document produces one class label

Use a separate test set to measure performance. The tiny example shows how the two processing steps connect.