Sentiment Analysis Tool

Implements sentiment analysis pipelines for extracting opinions and emotional tone from text data. Covers document-level and aspect-level sentiment, approach selection (lexicon-based, transformer models, LLM prompting), multilingual analysis, sarcasm and negation handling, temporal trend tracking, and visualization dashboards for business decision-making.

Usage

Describe your text data source (reviews, tweets, support tickets, survey responses), the granularity needed (overall sentiment vs aspect-level), languages involved, and how results will be used. Specify volume and latency requirements. This skill provides an implementation plan with model selection, preprocessing pipeline, and output format design.

Examples

• "Build a sentiment pipeline for analyzing 10K daily product reviews with aspect extraction for quality, shipping, and price"
• "Create an LLM-based sentiment analyzer for customer support emails that detects frustration level on a 1-5 scale"
• "Design a real-time social media sentiment tracker for brand mentions across Twitter and Reddit"

Guidelines

• Define your sentiment scale clearly: binary (positive/negative), ternary (+neutral), or fine-grained (1-5 stars)
• Use aspect-based sentiment when overall sentiment masks conflicting opinions about different features
• Preprocess text to handle emojis, slang, abbreviations, and platform-specific conventions
• Account for negation ("not bad" is positive) and sarcasm ("great, another bug") as major error sources
• Calibrate confidence scores so the model abstains on ambiguous text rather than guessing
• Benchmark against human annotator agreement (inter-rater reliability) as the accuracy ceiling
• Track sentiment trends over time windows rather than reacting to individual data points
• Segment results by customer demographics or product lines for actionable business insights