Sculley, D., Holt, G., Golovin, D., Davydov, E., Phillips, T., Ebner, D., … & Dennison, D. (2015). Hidden technical debt in machine learning systems. Advances in neural information processing systems, 28.
This paper presents a method for learning embeddings for text, particularly focusing on unsupervised learning from large text corpora. It introduces the Paragraph Vector algorithm, which aims to represent variable-length pieces of text, such as sentences, paragraphs, or documents, in a fixed-length vector. The approach involves training a model to predict words within the context of a paragraph, allowing the learned vectors to capture the semantic meaning of the text. The results demonstrate improved performance in tasks like sentiment analysis and information retrieval compared to traditional bag-of-words models.