Amazon Reviews using NLP (Tensorflow)

Natural Language Processing (NLP) is used to analyze and summarize Amazon reviews based on recent studies and comments

Amazon hosts millions of product reviews, which contain rich insights into customer satisfaction, product quality, and user experience. NLP helps automate the understanding of this massive data, enabling:

• Better product recommendations

• Smarter inventory and marketing decisions

• Enhanced customer feedback analysis

Step 1: Importing the required packages

import tensorflow as tf
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import nlp
import random

Step 2: Defining the function needed in the beginning itself

def show_history(h):
    epochs_trained = len(h.history['loss'])
    plt.figure(figsize=(16, 6))

    plt.subplot(1, 2, 1)
    plt.plot(range(0, epochs_trained), h.history.get('accuracy'), label='Training')
    plt.plot(range(0, epochs_trained), h.history.get('val_accuracy'), label='Validation')
    plt.ylim([0., 1.])
    plt.xlabel('Epochs')
    plt.ylabel('Accuracy')
    plt.legend()

    plt.subplot(1, 2, 2)
    plt.plot(range(0, epochs_trained), h.history.get('loss'), label='Training')
    plt.plot(range(0, epochs_trained), h.history.get('val_loss'), label='Validation')
    plt.xlabel('Epochs')
    plt.ylabel('Loss')
    plt.legend()
    plt.show()


def show_confusion_matrix(y_true, y_pred, classes):
    from sklearn.metrics import confusion_matrix

    cm = confusion_matrix(y_true, y_pred, normalize='true')

    plt.figure(figsize=(8, 8))
    sp = plt.subplot(1, 1, 1)
    ctx = sp.matshow(cm)
    plt.xticks(list(range(len(classes))), labels=classes)
    plt.yticks(list(range(len(classes))), labels=classes)
    plt.colorbar(ctx)
    plt.show()


print('Using TensorFlow version', tf.__version__)

Step 3: Importing the dataset and using the built-in dataset for practising NLP

amazon_data = pd.read_csv('/content/amazon_reviews.csv', header=None)
amazon_data.head()

Step 3.1: Perform the exploratory data analysis

new_header = amazon_data.iloc[0]
amazon_data = amazon_data[1:]
amazon_data.columns = new_header
amazon_data.head()
reviewId	userName	content	score	thumbsUpCount	reviewCreatedVersion	at	appVersion
1	032bcafa-54e0-4eb5-9e43-c86f8c5ac5e0	Garry Gordon	perfect	5	0	30.14.0.100	2025-08-04 12:33:57	30.14.0.100
2	9109695d-0b7d-48aa-a5a8-232e3978c7be	Marcia Chapman	you make it EXTREMELY IMPOSSIBLE to find and p...	1	0	30.14.0.100	2025-08-04 11:46:54	30.14.0.100
3	d329bbac-b2e2-40a8-bcec-1dea9aa35233	DominicJ Romero	Thanks for sharing	4	0	30.14.0.100	2025-08-04 10:31:01	30.14.0.100
4	c07a7bd2-79f9-4fbf-8599-f7ba1eb0e867	John Hulme	bril	5	0	30.14.0.100	2025-08-04 09:35:44	30.14.0.100
5	5a600712-8e68-4334-9cfe-16f275d8ccfe	Royce Miller	there's no way to talk directly with an agent....	1	0	30.14.0.100	2025-08-04 09:30:15	30.14.0.100

amazon_data.columns

Index(['reviewId', 'userName', 'content', 'score', 'thumbsUpCount',
       'reviewCreatedVersion', 'at', 'appVersion'],
      dtype='object', name=0)

amazon_data.shape
(77129, 8)

Step 4: Separate test and training dataset(not used validation test intentionally)

amazon_data_train = amazon_data.sample(frac=0.8, random_state=25)

print("No. of training samples: ", amazon_data_train.shape[0])
No. of training samples:  61703

amazon_data_test = amazon_data.drop(amazon_data_train.index)

print("No. of testing samples: ", amazon_data_test.shape[0])

No. of testing samples:  15426

amazon_data_train['content'].head()

content
27493	The app is ok and to be honest the customer se...
29435	Pretty much done. As Amazon continues to add f...
67392	Amazon has always delivered on time. Returns a...
64456	The crashes I'm unable to open it since yester...
63281	Everytime I search for any book the first opti...

dtype: object

amazon_data_test['content'].head()

content
21	Shes my mistress for christmas
22	buy a lot ,very usefull
23	very easy & convenient
30	Easy ordering and many choices
50	good morning

dtype: object

def get_review(data):
  reviews = [x['text'] for x in data]
  return reviews

train_amazon_reviews = amazon_data_train['content']
test_amazon_reviews = amazon_data_test['content']

train_amazon_reviews.iloc[20]

I HATE your new filter. It's slow. It's Not intuitive. It's medevil. It doesn't produce results any better than if i did it on my own. Your search results typically are not even relevant to what I'm looking for. I could pay the neighbor kid five bucks to walk to the store and buy what I need faster. I used to spend 75% of my shopping budget with Amazon but now, I don't even spend 10%. Not the Amazon cares because they aren't responding or doing anything to change.

test_amazon_reviews.iloc[30]
Enjoy using app

Step 5: Create Tokenizer

from tensorflow.keras.preprocessing.text import Tokenizer

tokenizer = Tokenizer(num_words=10000, oov_token='<UNK>')
tokenizer.fit_on_texts(train_amazon_reviews.fillna(''))

train_amazon_reviews.iloc[0]

The app is ok and to be honest the customer service is great, my issue is the delivery drivers. So many times a package gets delivered to a neighbour or somewhere completely different even though I have put a full description of where my property is and how to get to it, very poor service.

lengths_train = [t.split() for t in train_amazon_reviews.fillna('')]
plt.hist([len(t) for t in lengths_train])
plt.show()

max_length = 130

from tensorflow.keras.preprocessing.sequence import pad_sequences

def get_sequences(tokenizer, reviews):
  sequences = tokenizer.texts_to_sequences(reviews)
  padded_sequences = pad_sequences(sequences, truncating='post', maxlen=max_length)
  return padded_sequences

padded_train_amazon_reviews = get_sequences(tokenizer, train_amazon_reviews.fillna(''))
padded_test_amazon_reviews = get_sequences(tokenizer, test_amazon_reviews.fillna(''))

padded_train_amazon_reviews[0]

array([   0,    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,
          0,    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,
          0,    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,
          0,    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,
          0,    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,
          0,    0,    0,    0,    0,    0,    0,    0,    0,    0,    0,
          0,    0,    0,    0,    0,    0,    0,    0,    0,    2,    8,
          9,  526,    5,    3,   30, 1116,    2,   60,   46,    9,   64,
         11,  182,    9,    2,   52,  486,   25,  154,  130,    7,  181,
        358,  144,    3,    7, 4651,   31, 1055,  364,  253,   57,  328,
          4,   17,  235,    7,  481, 1060,   13,  191,   11, 2967,    9,
          5,  128,    3,   27,    3,    6,   59,  427,   46], dtype=int32)

padded_test_amazon_reviews[0]

array([  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
         0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,
         0,   0,   0,   0,   0,   0,   0,   0,   1,  11,   1,  12, 666],
      dtype=int32)

Step 6: Create the labels

train_classes = amazon_data_train['score']
test_classes = amazon_data_test['score']

train_classes

score
27493	2
29435	1
67392	5
64456	1
63281	2
...	...
29209	1
8733	4
64237	1
27272	2
41266	1
61703 rows × 1 columns


dtype: object

test_classes

score
21	5
22	5
23	5
30	5
50	5
...	...
77082	5
77098	1
77105	1
77117	1
77120	1
15426 rows × 1 columns


dtype: object

train_classes_numeric = pd.to_numeric(train_classes, errors='coerce').dropna()
plt.hist(train_classes_numeric)
plt.show()

test_classes_numeric = pd.to_numeric(test_classes, errors='coerce').dropna()
plt.hist(test_classes_numeric)
plt.show()

class_to_index = dict((c, i) for i, c in enumerate(set(train_classes_numeric)))
class_to_index

{1: 0, 2: 1, 3: 2, 4: 3, 5: 4}

index_to_class = dict((v, k) for k, v in class_to_index.items())
index_to_class

{0: 1, 1: 2, 2: 3, 3: 4, 4: 5}

Step 7: Create and Train the model

model = tf.keras.models.Sequential([
    tf.keras.layers.Embedding(10000, 16, input_length=max_length),
    tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(20, return_sequences=True)),
    tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(20)),
    tf.keras.layers.Dense(5, activation='softmax') # Change to 5 units and softmax for multi-class
])

model.compile(
    loss='sparse_categorical_crossentropy', # Change loss for integer labels
    optimizer='adam',
    metrics=['accuracy']
)

model.summary()

Model: "sequential_2"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ embedding_2 (Embedding)         │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ bidirectional_4 (Bidirectional) │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ bidirectional_5 (Bidirectional) │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_2 (Dense)                 │ ?                      │   0 (unbuilt) │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 0 (0.00 B)

 Trainable params: 0 (0.00 B)

 Non-trainable params: 0 (0.00 B)

val_reviews_train = amazon_data_train['score']

val_reviews_train.head()

score
27493	2
29435	1
67392	5
64456	1
63281	2

dtype: object

val_sequence_train = get_sequences(tokenizer, train_amazon_reviews.fillna(''))

val_reviews_train.iloc[0],val_labels_train[0]

h = model.fit(
    padded_train_amazon_reviews,
    val_labels_train,
    validation_data=(padded_test_amazon_reviews, test_labels), # Use test_labels instead of test_classes_numeric
    epochs=10,
    callbacks= [tf.keras.callbacks.EarlyStopping(monitor='val_accuracy', patience=2)]
)

show_history(h)

Step8 : Evaluate the Model and Perform a Confusion Matrix

test_reviews = amazon_data_test['content']
test_sequences = get_sequences(tokenizer, test_reviews.fillna(''))
test_labels = np.array([class_to_index[score] for score in test_classes_numeric])

_= model.evaluate(test_sequences, test_labels, verbose=0)

i = random.randint(0, len(test_reviews) - 1)

print("Review:", test_reviews.iloc[i])
print("Label:", test_labels[i])

p = model.predict(np.expand_dims(test_sequences[i], axis=0))[0]

Review: Amazon is good & fast delivery
Label: 4
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 57ms/step

p = model.predict(single_test_sequence)[0]

print("Predicted probabilities:", p)

predicted_class = np.argmax(p)
print("Predicted class:", predicted_class)

Predicted class: 4

show_confusion_matrix(test_labels, np.argmax(model.predict(test_sequences), axis=1), list(class_to_index.keys()))