Use a vector embedding technique to create a vector embedding for input text. You can choose a technique based on the pre-trained model you want to use to convert the text to a vector.
A vector embedding represents the text as an array of numbers. Each element in the array represents a different dimension of the text. To create vector embeddings, select an input column for embedding and then select one of the following vector embedding techniques:
Word embedding
Convert each word to a vector using the Word2Vec Gigaword 5th Edition model with 300 dimensions (word2vec_giga_300). Useful for text classification and sentiment analysis.
BERT embedding
Convert each sentence to a vector using the Smaller BERT Embedding (L-2_H-768_A-12) model with 768 dimensions (small_bert_L2_768). Useful for text classification and semantic search.
Consider the following rules and guidelines for vector embeddings:
•Vector embeddings created by different embedding models can't be compared even if they have the same dimensions. If you switch between embedding models, rerun the mapping, including all Source, Chunking, Vector Embedding, and Target transformations, to create embeddings for all documents using the new model.
•Because the Vector Embedding transformation is a passive transformation that produces one output row for each input row, input columns that contain null or empty strings return an empty output vector. If the vector is empty, vector databases like Pinecone might drop the row.
•The Vector Embedding transformation can process only English text.