MLA-C01 · Question #119
An insurance company needs to automate claim compliance reviews because human reviews are expensive and error-prone. The company has a large set of claims and a compliance label for each. Each claim…
The correct answer is D. Apply Amazon SageMaker Object2Vec to claims in the training set. Send the derived features. Option D is correct because Amazon SageMaker Object2Vec is specifically designed to encode complex objects - including multi-sentence documents - into dense, fixed-length embedding vectors that capture semantic meaning. These embeddings serve as rich feature representations…
Question
An insurance company needs to automate claim compliance reviews because human reviews are expensive and error-prone. The company has a large set of claims and a compliance label for each. Each claim consists of a few sentences in English, many of which contain complex related information. Management would like to use Amazon SageMaker built-in algorithms to design a machine learning supervised model that can be trained to read each claim and predict if the claim is compliant or not. Which approach should be used to extract features from the claims to be used as inputs for the downstream supervised task?
Options
- ADerive a dictionary of tokens from claims in the entire dataset. Apply one-hot encoding to tokens
- BApply Amazon SageMaker BlazingText in Word2Vec mode to claims in the training set. Send the
- CApply Amazon SageMaker BlazingText in classification mode to labeled claims in the training set
- DApply Amazon SageMaker Object2Vec to claims in the training set. Send the derived features
How the community answered
(32 responses)- A3% (1)
- B13% (4)
- C6% (2)
- D78% (25)
Explanation
Option D is correct because Amazon SageMaker Object2Vec is specifically designed to encode complex objects - including multi-sentence documents - into dense, fixed-length embedding vectors that capture semantic meaning. These embeddings serve as rich feature representations suitable for feeding into a downstream supervised classifier, which is exactly the two-stage pipeline the question describes.
Why the distractors are wrong:
- A (One-hot encoding): Produces high-dimensional, sparse vectors with no semantic meaning - word order and context are lost entirely, making it poorly suited for complex insurance language.
- B (BlazingText Word2Vec mode): Word2Vec generates word-level embeddings, not document- or sentence-level embeddings. You'd have no clean way to aggregate word vectors into a single claim-level feature vector for classification.
- C (BlazingText classification mode): This is an end-to-end text classifier (similar to fastText) - it is the downstream task, not a feature extractor. Using it contradicts the question's requirement to extract features for a separate supervised model.
Memory tip: Think of it this way - Word2Vec embeds words, BlazingText classification is the classifier, and Object2Vec embeds objects (sentences, documents, pairs). When the question says "extract features for a downstream task," that's Object2Vec's exact job description.
Topics
Community Discussion
No community discussion yet for this question.