NCA-GENL Question 53
Single answerYou are developing a recommendation system for a retail platform and need to create text embeddings for product descriptions to measure similarity between them. Which type of model or approach would be most suitable for generating high-quality text embeddings in this context?
- A
A pre-trained transformer-based model like BERT or OpenAI’s GPT
- B
A simple bag-of-words model
- C
A rule-based natural language processing pipeline
- D
A convolutional neural network (CNN) trained on image datasets
Show answer and explanation
Correct answer: A
Explanation
The best approach for creating text embeddings in this scenario is to use a pre-trained transformer-based model, such as BERT or GPT. These models generate embeddings that capture contextual and semantic relationships, making them ideal for tasks like calculating similarity between product descriptions. Other methods, like bag-of-words or rule-based pipelines, lack the ability to encode rich contextual information, and CNNs are not applicable for text data.
- A. Correct.
Transformer-based models like BERT or GPT are well-suited for generating high-quality text embeddings due to their ability to capture contextual relationships and semantic meaning in text.
- B. Incorrect.
A bag-of-words model is a basic approach that does not capture contextual relationships between words, leading to less effective embeddings compared to transformer-based models.
- C. Incorrect.
A rule-based natural language processing pipeline is not designed for generating embeddings and is typically used for specific linguistic tasks like parsing or entity extraction.
- D. Incorrect.
A convolutional neural network (CNN) trained on image datasets is unsuitable for text tasks as it is primarily designed for image feature extraction, not text embeddings.