Google Professional Machine Learning Engineer Question 74
Select 3Google Cloud PlatformYou are a Machine Learning Engineer working for an e-commerce company. The company has tabular data containing customer demographics, product details, and purchase history. They also have thousands of customer reviews available in text format. Your task is to create a custom model to predict the likelihood of a customer making a purchase. Which of the following approaches should you prioritize to effectively utilize the available data?
- A
Use a feature engineering pipeline to preprocess the tabular data and train a gradient-boosted decision tree model.
- B
Convert the customer review text into numerical embeddings using a pre-trained natural language processing (NLP) model and incorporate these embeddings into your training data.
- C
Discard the customer review text data and focus on training a tabular data model, as text data is harder to process.
- D
Use a multi-modal model that combines both tabular data and text data to predict the likelihood of a customer purchase.
- E
Train separate models for text data and tabular data, and combine their predictions using an ensemble method.
Show answer and explanation
Correct answers: A, B, D
Explanation
To effectively utilize the available data, it's important to preprocess and use both tabular and text data. Tabular data can be processed effectively using models like gradient-boosted decision trees, while text data can be converted into numerical embeddings using pre-trained NLP models. Combining these modalities in a multi-modal model allows you to leverage the strengths of both data types for better predictive performance.
- A. Correct.
Correct: Preprocessing tabular data is essential for training models like gradient-boosted decision trees, which excel in handling structured data.
- B. Correct.
Correct: Text embeddings from pre-trained NLP models (e.g., BERT or Word2Vec) are highly effective in leveraging textual data for predictive tasks.
- C. Incorrect.
Incorrect: Discarding text data would result in lost information that could be valuable for the prediction task, especially since customer reviews might contain sentiment or intent signals.
- D. Correct.
Correct: Multi-modal models efficiently combine data from different modalities, such as tabular and text data, to improve prediction accuracy.
- E. Incorrect.
Incorrect: While ensemble methods can be effective, training separate models for tabular and text data may increase complexity without guaranteeing better performance compared to a unified multi-modal approach.