all-minilm-l6-v2-civic

This is a sentence-transformers model fine-tuned using TSDAE (Transformer-based Denoising AutoEncoder) on government/civic meeting transcripts.

Model Details

• Base Model: all-minilm-l6-v2
• Training Method: TSDAE (unsupervised domain adaptation)
• Original Source: ar9av/all-minilm-l6-v2-civic

Description

TSDAE fine-tuned model on civic/government meeting transcripts. The model was trained on 10,000 government meeting transcripts to learn domain-specific representations while remaining general-purpose for downstream tasks.

Key Features

• ✅ Learned domain-specific abbreviations (BOS = Board of Supervisors, etc.)
• ✅ Enhanced understanding of government/civic terminology
• ✅ General-purpose embeddings suitable for various downstream tasks

Usage

from sentence_transformers import SentenceTransformer

model = SentenceTransformer('TSG/all-minilm-l6-v2-civic')
embeddings = model.encode(['Your text here'])

Training Data

• 10,000 government/civic meeting transcripts
• Various meeting types: City Council, Planning Commission, Board of Education, etc.
• Text cleaned and chunked for training

Evaluation

The model shows significant improvement in:

• Domain-specific semantic similarity
• Abbreviation understanding (BOS, CC, BOE, etc.)
• Clustering quality for civic domain

Intended Use

This model is designed for general-purpose semantic similarity and embedding tasks, with enhanced understanding of government/civic domain language.

Original Model

This model was originally trained and hosted at: https://huggingface.co/ar9av/all-minilm-l6-v2-civic