Apparel Fit Sentiment Analysis

Analyze customer reviews for sizing and fit feedback

Model Selection

Choose whether to use the built-in rule-based SpaCy + VADER model, or a BERT model loaded from local directories on this machine.

Analyze a Single Review

Batch Analysis from JSON

Upload a JSON file containing a list of review strings. View expected format

Confusion Matrix & Performance Metrics

Upload a JSON file with test data containing reviews and ground truth labels. View expected format

SpaCy + VADER rule-based model

When you select SpaCy + VADER, the model:

Splits the review into windows using simple conjunctions like and, but, and commas.
For each aspect, focuses on windows that mention that aspect (or the whole review if none do).
Looks for small, loose, and perfect fit keywords in each window.
Uses VADER sentiment (from −1 to 1) to decide whether those keywords indicate Small/Tight, Perfect/Fit, or Large/Loose.
Aggregates votes across windows; if there is no consistent signal, the prediction is Irrelevant.

Extractor Performance

Upload a JSON file with reviews and labeled aspects to evaluate the aspect extractor. View expected format

SpaCy rule-based extractor

When you select SpaCy + VADER for the extractor, the model:

Parses the review with spaCy and finds noun phrases and NOUN/PROPN tokens.
Keeps only candidates that exactly match a curated list of aspect keywords (e.g. waist, bust, sleeves, shoulder straps).
Returns those matched phrases as the extracted aspects, without any learning.

Expected JSON Format

The JSON file should be a simple array of review strings:

[
  "The waist is too small but the length is perfect.",
  "I love the blue sleeves.",
  "The shoulder straps dug in a bit, and the hem was huge.",
  "Perfect fit overall, very comfortable."
]

Each string in the array is treated as a separate review to analyze.

Expected Extractor Test Data Format

JSON array of objects with "text" and "aspects":

[
  { "text": "The waist fits perfectly.", "aspects": ["waist"] },
  { "text": "Sleeves are way too long.", "aspects": ["sleeves"] },
  { "text": "Great dress but the bust is tight.", "aspects": ["dress", "bust"] },
  { "text": "The hemline drags on the floor.", "aspects": ["hemline"] },
  { "text": "Fabric is scratchy and the zipper is stuck.", "aspects": ["fabric", "zipper"] }
]

Aspects use exact match (e.g. "shoe lace" must be extracted as "shoe lace", not just "shoe").

Expected Test Data Format

The JSON file should be an array of objects with "text" and "labels" fields:

[
  { "text": "The dress fits perfectly.", "labels": [{ "aspect": "dress", "score": 1 }] },
  { "text": "Way too small in the bust.", "labels": [{ "aspect": "bust", "score": 0 }] },
  { "text": "The sleeves are huge and baggy.", "labels": [{ "aspect": "sleeves", "score": 2 }] },
  { "text": "I love the blue color.", "labels": [{ "aspect": "color", "score": 3 }] },
  { "text": "The waist was tight but the length was perfect.", 
    "labels": [{ "aspect": "waist", "score": 0 }, { "aspect": "length", "score": 1 }] }
]

Score values: 0 = Small/Tight, 1 = Perfect/Fit, 2 = Large/Loose, 3 = Irrelevant
Each review can have multiple aspect-label pairs.

Apparel Fit Sentiment Analysis

Model Selection

Analyze a Single Review

Analysis Results

Batch Analysis from JSON

Aggregated Analysis Results

Confusion Matrix & Performance Metrics

SpaCy + VADER rule-based model

Model Performance Metrics

Extractor Performance

SpaCy rule-based extractor

Extractor Metrics

Review-by-Review: Predicted vs Labeled

Model Selection

Analyze a Single Review

Analysis Results

Batch Analysis from JSON

Aggregated Analysis Results

Confusion Matrix & Performance Metrics

SpaCy + VADER rule-based model

Model Performance Metrics

Extractor Performance

SpaCy rule-based extractor

Extractor Metrics

Review-by-Review: Predicted vs Labeled

Expected JSON Format

Expected Extractor Test Data Format

Expected Test Data Format

Reviews