Text analysis computationally examines written content to find patterns, themes, frequencies, and other useful properties. It turns large volumes of messy, unstructured material—such as reviews, emails, and notes—into evidence for business decisions, without requiring someone to read everything manually. section 5.0
Common text-analysis tasks include:
- Perform sentiment analysis to understand expressed opinions. section 2.6
- Use topic modeling to identify recurring subjects and issues. section 2.6
- Count words or phrases to reveal frequent themes and patterns. section 2.6
- Clean spelling, punctuation, capitalization, headers, metadata, abbreviations, slang, jargon, and other noise before interpreting results. section 5.0
Customer-review analysis
A customer-facing team can examine review titles and messages alongside review scores, dates, and response times. Text findings can expose common complaints or themes, while raw comments and sample cases should be checked manually before decisions rely on them. section 5.2 section 2.6
DataFrame name: df_order_reviews Columns: - review_id (object): Unique identifier of each review. - order_id (object): Identifier of the order being reviewed. - review_score (int64): Note ranging from 1 to 5 given by the customer on a satisfaction survey. - review_comment_title (object): Comment title from the review left by the customer in Portuguese. - review_comment_message (object): Comment message from the review left by the customer in Portuguese. - review_creation_date (datetime64[ns]): The timestamp of the review creation. - review_answer_timestamp (datetime64[ns]): I am not sure what this one actually is.
Inspect the review dataset structure, including Portuguese comment titles and messages, scores, identifiers, and timestamps. section 2.6
Generative AI can support information gathering, analytical planning, interpretation, coding, and quick initial analyses through direct interaction or an API. Dedicated libraries may provide greater accuracy for specialized vocabulary, ambiguity, consistency, speed, and scalability, so outputs require validation. section 5.0 section 5.1.3
