14. Natural Language Insights

Local LLMs extract meaningful insights from unstructured text data without sending sensitive information externally.

Sentiment Analysis with Local Models

import ollama

def analyze_sentiment(text, model='llama3.2'):
    prompt = f"""Analyze the sentiment of this text. Return ONLY a single word:
- "positive" for positive sentiment
- "negative" for negative sentiment  
- "neutral" for neutral sentiment

Text: {text}"""
    
    response = ollama.chat(model=model, messages=[
        {'role': 'user', 'content': prompt}
    ])
    return response['message']['content'].strip().lower()

# Batch process reviews
df['sentiment'] = df['review_text'].apply(lambda x: analyze_sentiment(str(x)[:500]))

Topic Extraction

def extract_topics(texts, n_topics=5):
    prompt = f"""From these texts, identify {n_topics} main topics. 
    Return a comma-separated list of topic names.
    
    Texts:
    {texts[:2000]}"""
    
    response = ollama.chat(model='llama3.2', messages=[
        {'role': 'user', 'content': prompt}
    ])
    topics = [t.strip() for t in response['message']['content'].split(',')]
    return topics

corpus = ' '.join(df['comments'].dropna().tolist())
main_topics = extract_topics(corpus, n_topics=5)

Named Entity Recognition (NER)

Extract entities from text using structured prompting:

def extract_entities(text):
    prompt = f"""Extract named entities from this text. Format as:
Entity | Type (ORGANIZATION, PERSON, LOCATION, DATE, PRODUCT)

Text: {text}

Only include actual entities found in the text."""

    response = ollama.chat(model='llama3.2', messages=[
        {'role': 'user', 'content': prompt}
    ])
    return response['message']['content']

sample_text = "Apple released iPhone 15 on September 12, 2023 at their Cupertino headquarters."
entities = extract_entities(sample_text)
print(entities)
# Output: Apple | ORGANIZATION
#         iPhone 15 | PRODUCT  
#         September 12, 2023 | DATE
#         Cupertino | LOCATION

Text Summarization

def summarize_text(text, max_length=200):
    prompt = f"""Summarize this text in no more than {max_length} characters.
    Include the main points and key numbers if present.

    Text: {text}"""

    response = ollama.chat(model='llama3.2', messages=[
        {'role': 'user', 'content': prompt}
    ])
    return response['message']['content']

# Summarize customer feedback themes
feedback_corpus = df['feedback'].dropna().head(100).tolist()
summary = summarize_text(' '.join(feedback_corpus))

Question Answering from Data

def answer_question(documents, question):
    prompt = f"""Based on the following documents, answer the question.
    If the answer is not in the documents, say "I don't know".

    Documents:
    {documents}

    Question: {question}"""

    response = ollama.chat(model='llama3.2', messages=[
        {'role': 'user', 'content': prompt}
    ])
    return response['message']['content']

# Query your data
relevant_docs = df[df['category'] == 'technical_issue']['description'].head(10).tolist()
answer = answer_question(relevant_docs, "What are the most common technical issues?")