Amazon Comprehend Positivity Score

The Amazon Comprehend Positivity Score is a sentiment analysis metric provided by Amazon Comprehend, a natural language processing (NLP) service offered by AWS. It quantifies how positive the sentiment of a given text is.

What is it exactly?

  • Amazon Comprehend analyzes input text and returns a sentiment classification (e.g., POSITIVE, NEGATIVE, NEUTRAL, or MIXED).

  • Alongside the classification, it provides sentiment scores for each category.

  • The Positivity Score (often called Positive Score) is a decimal value between 0 and 1 that represents the confidence or strength of the positive sentiment in the text.

How to interpret the Positivity Score?

  • Close to 1: The text is strongly positive.

  • Around 0.5: The text has a mixed or neutral sentiment with some positive aspects.

  • Close to 0: The text has little or no positive sentiment.

Example output snippet from Amazon Comprehend Sentiment Analysis API:

json

Copy

{ "Sentiment": "POSITIVE", "SentimentScore": { "Positive": 0.98, "Negative": 0.01, "Neutral": 0.005, "Mixed": 0.005 } }

Here, the Positive score is 0.98, indicating a very positive text.

If you want me to help you with how to call Amazon Comprehend API to get this score or how to interpret or use it in an application, just ask!

You want to understand how to analyze sentiment at different levels

You want to understand how to analyze sentiment at different levels — brand, product category, and specific product SKUs — across social media and community platforms like Reddit. Here’s a structured approach to doing that using Amazon Comprehend (and similar tools):

1. Data Collection

First, gather the text data where people discuss your brand, product categories, and SKUs. For social media and communities like Reddit, you can:

  • Use APIs or web scraping tools (e.g., Reddit API, Twitter API, social listening tools like Brandwatch, Sprout Social).

  • Collect posts, comments, reviews, and discussions mentioning:

    • Brand name(s)

    • Product category keywords (e.g., “running shoes,” “smartphones”)

    • Specific product SKUs or model names

2. Text Preprocessing & Entity Extraction

Before sentiment analysis, preprocess and organize data:

  • Clean text: Remove URLs, emojis (or convert them to text), special characters.

  • Identify mentions:

    • Detect whether a comment refers to brand-level, category-level, or SKU-level.

    • Use Named Entity Recognition (NER) to extract entities like brand names, product types, and SKUs.

    • Amazon Comprehend offers built-in entity recognition you can customize or extend with custom entities for SKU detection.

3. Sentiment Analysis per Text Snippet

Run sentiment analysis on each text snippet (post/comment) using Amazon Comprehend or similar NLP services:

  • You get a sentiment label and sentiment scores (positive, negative, neutral, mixed).

  • Store the sentiment result linked to the detected entity (brand, category, SKU).

4. Aggregation & Scoring

Aggregate the sentiment results at each level:

LevelWhat to AggregateExample MetricBrandAll texts mentioning the brandAverage positivity score, sentiment distribution (%)CategoryTexts mentioning product categoriesAverage sentiment, volume of mentionsSKUTexts mentioning specific SKUs or product modelsSKU-level sentiment trends, sentiment volatility

You can calculate:

  • Positivity score averages over a time window

  • Sentiment trends over time (daily, weekly)

  • Sentiment distribution: % positive vs negative mentions

  • Volume of mentions alongside sentiment to understand impact

5. Visualization & Insights

Use dashboards to visualize:

  • Sentiment heatmaps by brand, category, SKU

  • Time series showing how sentiment evolves

  • Sentiment vs mention volume correlation

  • Highlight spikes of negative sentiment for early issue detection

6. Advanced Analysis & Context

  • Topic modeling to see what issues or features are driving sentiment

  • Use aspect-based sentiment analysis to detect sentiment about specific product features (battery life, customer service, price)

  • Combine with influencer or community user analysis to identify influential posts driving sentiment

How Amazon Comprehend Fits In

  • Use Amazon Comprehend for entity detection, sentiment analysis, and optionally key phrase extraction.

  • You can build a pipeline:

    1. Pull social data from Reddit API, Twitter API, etc.

    2. Clean and preprocess text.

    3. Call Amazon Comprehend for NER + sentiment.

    4. Store results in a database with mappings to brand/category/SKU.

    5. Aggregate and analyze sentiment data.

Bonus: Handling Reddit and Communities

  • Reddit posts and comments often have slang, sarcasm, and complex context. Consider:

    • Supplementing Amazon Comprehend with custom models trained on Reddit data.

    • Using context-aware models or fine-tuning open-source models (e.g., Hugging Face transformers) specialized for social media language.

    • Incorporate metadata (e.g., subreddit topic, user karma) for weighting sentiment relevance.

Amazon, Amazon Comprehend, SentimentFrancesca Tabor