Exam AWS Certified Machine Learning - Specialty topic 1 question 20 discussion

the solution is word embedding. As it is a interactive online dictionary, we need pre-trained word embedding thus the answer is D. In addition, there is no mention that the online dictonary is unique and does not have a pre-trained word embedding. Thus I strongly feel the answer is D

D is correct. It is not a specialized dictionary so use the existing word corpus to train the model

D. Download word embeddings pre-trained on a large corpus. Reason : For a nearest neighbor model that finds words used in similar contexts, word embeddings are the best choice. Pre-trained word embeddings capture semantic relationships and contextual similarity between words based on a large text corpus (e.g., Wikipedia, Common Crawl). The Specialist should: Use pre-trained word embeddings like Word2Vec, GloVe, or FastText. Load the embeddings into the model for efficient similarity comparisons. Use a nearest neighbor search algorithm (e.g., FAISS, k-d tree, Annoy) to quickly find similar words.

D. Download word embeddings pre-trained on a large corpus. Word embeddings are a type of dense representation of words, which encode semantic meaning in a vector form. These embeddings are typically pre-trained on a large corpus of text data, such as a large set of books, news articles, or web pages, and capture the context in which words are used. Word embeddings can be used as features for a nearest neighbor model, which can be used to find words used in similar contexts. Downloading pre-trained word embeddings is a good way to get started quickly and leverage the strengths of these representations, which have been optimized on a large amount of data. This is likely to result in more accurate and reliable features than other options like one-hot encoding, edit distance, or using Amazon Mechanical Turk to produce synonyms.

A. NO - one-hot encoding is a very early featurization stage B. NO - we don't want human labelling C. NO - too costly to do from scratch D. YES - leverage exiting training; the word embeddings will provide vectors than be used to measure distance in the downstream nearest neighbor model

Pre-trained word embeddings, such as Word2Vec, GloVe, or FastText, capture the semantic and contextual meaning of words based on a large corpus of text data. By downloading pre-trained word embeddings, the Specialist can leverage the semantic relationships between words to provide meaningful word features for the nearest neighbor model powering the widget. Utilizing pre-trained word embeddings allows the model to understand and display words used in similar contexts effectively.

A. One-hot word encoding vectors: These vectors represent words by marking them as present or absent in a fixed-length binary vector. However, they don't capture relationships between words or their meanings. B. Producing synonyms: This would involve generating similar words for each word manually, which could be time-consuming and might not cover all possible contexts. C. Word embedding vectors based on edit distance: This approach focuses on how similar words are in terms of their spelling or characters, not necessarily their meaning or context in sentences. D. Downloading pre-trained word embeddings: These are vectors that represent words based on their contextual usage in a large dataset, capturing relationships between words and their meanings.

correct D ay tupoy

words that are used in similar contexts will have vectors that are close in the embedding space

D is correct

I also believe that D is the correct answer. No reason to create word embeddings from scratch

1. One-hot encoding will blow up the feature space - it is not recommended for a high cardinality problem domain. 2. One still needs to train the word features on large bodies of text to map context to each word

12-sep exam

DDDDDDDDDDDDD

D for sure

Definitely D.

A)It requires that document text be cleaned and prepared such that each word is one-hot encoded. Ref:https://machinelearningmastery.com/what-are-word-embeddings/