Named entity recognition in NLP - ML Experiment: Train & Evaluate

Experiment - Named entity recognition

Problem:We want to teach a computer to find names of people, places, and organizations in sentences. Our current model learns well on training data but does not do well on new sentences.

Current Metrics:Training accuracy: 95%, Validation accuracy: 70%

Issue:The model is overfitting: it performs very well on training data but poorly on validation data.

Your Task

Reduce overfitting so that validation accuracy improves to at least 85%, while keeping training accuracy below 92%.

You can only change the model architecture and training settings.

Do not change the dataset or add more data.

Hint 1

Hint 2

Hint 3

Solution

NLP

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Embedding, Bidirectional, LSTM, Dense, Dropout
from tensorflow.keras.callbacks import EarlyStopping

# Sample data placeholders (replace with actual data loading)
X_train = ...  # tokenized and padded sequences for training
y_train = ...  # one-hot encoded labels for training
X_val = ...    # tokenized and padded sequences for validation
y_val = ...    # one-hot encoded labels for validation

vocab_size = 10000  # example vocabulary size
embedding_dim = 64
max_len = 100  # max sequence length
num_classes = 10  # number of entity classes including 'O'

model = Sequential([
    Embedding(vocab_size, embedding_dim, input_length=max_len),
    Bidirectional(LSTM(64, return_sequences=True)),
    Dropout(0.5),
    Bidirectional(LSTM(32, return_sequences=True)),
    Dropout(0.5),
    Dense(num_classes, activation='softmax')
])

model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

early_stop = EarlyStopping(monitor='val_loss', patience=3, restore_best_weights=True)

history = model.fit(
    X_train, y_train,
    epochs=20,
    batch_size=32,
    validation_data=(X_val, y_val),
    callbacks=[early_stop]
)

# After training, evaluate on validation set
val_loss, val_accuracy = model.evaluate(X_val, y_val)
print(f'Validation accuracy: {val_accuracy * 100:.2f}%')

Added Dropout layers after LSTM layers to reduce overfitting.

Reduced the number of units in the second LSTM layer from 64 to 32 to lower model complexity.

Added EarlyStopping callback to stop training when validation loss stops improving.

Results Interpretation

Before: Training accuracy: 95%, Validation accuracy: 70% (overfitting)

After: Training accuracy: 90%, Validation accuracy: 87% (better generalization)

Adding dropout and early stopping helps the model generalize better by preventing it from memorizing training data, which reduces overfitting and improves validation accuracy.

Bonus Experiment

Try using a pre-trained language model like BERT for named entity recognition to see if it improves accuracy further.

💡 Hint

Use a library like Hugging Face Transformers to load a pre-trained BERT model and fine-tune it on your NER dataset.

Practice

1. What is the main goal of Named Entity Recognition (NER) in natural language processing?

easy

A. To find and label names of people, places, and dates in text

B. To translate text from one language to another

C. To summarize long documents into short paragraphs

D. To generate new text based on input

Solution

1. Step 1: Understand NER purpose

   NER is designed to identify and label specific types of information like names, places, and dates in text.

2. Step 2: Compare with other NLP tasks

   Translation, summarization, and text generation are different tasks unrelated to labeling entities.

3. Final Answer:

   To find and label names of people, places, and dates in text -> Option A

4. Quick Check:

   NER = Labeling names in text [OK]

Hint: NER finds names and dates in text, not translations or summaries [OK]

Common Mistakes:

• Confusing NER with translation or summarization
• Thinking NER generates new text
• Believing NER only finds keywords, not entities

2. Which of the following is the correct way to import a Named Entity Recognition pipeline using Hugging Face Transformers in Python?

easy

A. import pipeline from transformers; ner = pipeline('named_entity')

B. from transformers import pipeline; ner = pipeline('ner')

C. from transformers import ner_pipeline; ner = ner_pipeline()

D. import ner from transformers; ner = pipeline('ner')

Solution

1. Step 1: Recall correct import syntax

   The Hugging Face library uses 'from transformers import pipeline' to import the pipeline function.

2. Step 2: Check pipeline usage for NER

   Calling pipeline('ner') creates a named entity recognition pipeline correctly.

3. Final Answer:

   from transformers import pipeline; ner = pipeline('ner') -> Option B

4. Quick Check:

   Correct import and pipeline call = from transformers import pipeline; ner = pipeline('ner') [OK]

Hint: Use 'from transformers import pipeline' and call pipeline('ner') [OK]

Common Mistakes:

• Using incorrect import syntax
• Calling pipeline with wrong task name
• Trying to import non-existent functions

3. Given the following Python code using Hugging Face Transformers NER pipeline:

from transformers import pipeline
ner = pipeline('ner')
text = "Barack Obama was born in Hawaii on August 4, 1961."
results = ner(text)
print(results)

What will be the output type of results?

medium

A. A single string with all entities concatenated

B. A dictionary with entity counts

C. A list of dictionaries with entity details

D. An integer representing number of entities

Solution

1. Step 1: Understand pipeline output format

   The NER pipeline returns a list where each item is a dictionary describing an entity found in the text.

2. Step 2: Check example output structure

   Each dictionary contains keys like 'entity', 'score', 'index', and 'word' describing the entity.

3. Final Answer:

   A list of dictionaries with entity details -> Option C

4. Quick Check:

   NER output = list of entity dictionaries [OK]

Hint: NER pipeline returns list of dicts, not strings or counts [OK]

Common Mistakes:

• Expecting a single string output
• Thinking output is a dictionary summary
• Assuming output is just a count number

4. Consider this code snippet for NER using Hugging Face Transformers:

from transformers import pipeline
ner = pipeline('ner')
text = "Apple is looking at buying U.K. startup for $1 billion"
results = ner(text, grouped_entities=True)
print(results)

What is the likely error or issue here?

medium

A. The text input must be a list, not a string

B. The pipeline call is missing the model parameter

C. There is no error; code runs correctly

D. The argument 'grouped_entities' is invalid and causes a TypeError

Solution

1. Step 1: Check pipeline argument validity

   The 'grouped_entities' argument is not supported in the current pipeline call and will raise a TypeError.

2. Step 2: Confirm correct usage

   To group entities, the argument should be 'aggregation_strategy' with values like 'simple', not 'grouped_entities'.

3. Final Answer:

   The argument 'grouped_entities' is invalid and causes a TypeError -> Option D

4. Quick Check:

   Invalid argument causes error = The argument 'grouped_entities' is invalid and causes a TypeError [OK]

Hint: Check pipeline argument names carefully; 'grouped_entities' is wrong [OK]

Common Mistakes:

• Using unsupported argument names
• Assuming text input must be a list
• Thinking missing model parameter causes error

5. You want to extract all person names and locations from a news article using NER. Which approach best ensures you only get these two entity types from the pipeline output?

hard

A. Filter the NER results by checking if the entity label is 'PER' or 'LOC'

B. Use the pipeline with task='ner' and no filtering

C. Manually search the text for capitalized words

D. Train a new model only on person and location data

Solution

1. Step 1: Understand NER output labels

   NER results include entity labels like 'PER' for person and 'LOC' for location.

2. Step 2: Filter results for desired entities

   Filtering the output by these labels extracts only person and location entities effectively.

3. Final Answer:

   Filter the NER results by checking if the entity label is 'PER' or 'LOC' -> Option A

4. Quick Check:

   Filter by labels 'PER' and 'LOC' to get persons and locations [OK]

Hint: Filter NER output by entity labels to get specific types [OK]

Common Mistakes:

• Not filtering and getting all entity types
• Trying manual text search instead of using labels
• Assuming retraining is needed for filtering