NCA-GENL Question 51
Single answerYou are tasked with building a recommendation system for a movie streaming platform. The system needs to recommend movies based on user preferences and search history. To achieve this, you decide to create text embeddings for movie descriptions and user reviews. Which type of model would be most suitable for generating these embeddings?
- A
A pre-trained Large Language Model (LLM) like GPT, fine-tuned for text embedding tasks
- B
A convolutional neural network (CNN) designed for image classification
- C
A rule-based algorithm focused on keyword matching
- D
A specialized embedding model like Sentence Transformers or BERT trained for semantic similarity
Show answer and explanation
Correct answer: D
Explanation
Generating text embeddings for tasks like recommendation systems requires models capable of capturing semantic relationships in the text. Specialized models like Sentence Transformers or BERT are optimized for these tasks, providing efficient and high-quality embeddings suitable for semantic similarity and recommendation tasks. While pre-trained LLMs like GPT can be adapted, they are not as efficient as embedding-focused models.
- A. Incorrect.
Pre-trained LLMs like GPT can be adapted for text embeddings but are not as efficient or specialized for this task as embedding-focused models.
- B. Incorrect.
Convolutional neural networks (CNNs) are primarily designed for image data and are not suited for generating text embeddings.
- C. Incorrect.
Rule-based algorithms do not generate embeddings; they rely on keyword matching and lack the ability to capture semantic relationships in text.
- D. Correct.
Specialized embedding models like Sentence Transformers or BERT are designed to create high-quality text embeddings that capture semantic meaning, making them ideal for this task.