Document Embeddings

Based on the results of: https://arxiv.org/abs/2304.14796

Implemented Methods

• Average Pooling, with adjustable range for sentences used.
• PERT weighted average pooling

Usage

Wrapper for Sentence-Embedding, which is used to provide embedding functionality

from sentence_transformers import SentenceTransformer

sentence_model = SentenceTransformer("paraphrase-multilingual-MiniLM-L12-v2")

document_model = AverageDocumentEmbedding(sentence_model, language='german')


doc1 = "Arbitrary text"
doc2 = "..."

document_model.encode([doc1, doc2, ...])