Part 9/12:

Despite promising results, several limitations remain:

• Cold Start Problem: Newly launched products are not in the vocabulary, making it impossible for the model to recommend them initially.

• Resource Intensive: Larger models would demand significant computational resources, especially at scale.

• Limited External Factors: The current model relies solely on purchase sequences, ignoring broader influences like seasonality, promotions, demographics, and external trends.

Future Directions and Improvements

The researchers are exploring several enhancements:

• Incorporating product description embeddings using more sophisticated tokenizers (e.g., subword or semantic-based) to better capture product relationships.