Code for Transformers4Rec Next Course Recommender
Budget: €30 – €250 EUR
Project Overview:
I am looking for a complete implementation of a Transformers4Rec-based sequential next-course recommendation system. The system needs to work with my pre-made data splits to ensure rigorous hypothesis testing, and the splits are based on full user sequences. The project should include model selection and parameter tuning, considering my limited computational resources and the need to retrain the model multiple times.
Key Requirements:
- End-to-End Solution: From NVTabular schema creation to model training and evaluation.
- Metrics for Evaluation: Hit ratio at k, NDCG@k, Coverage (percentage of courses recommended), Mean - Reciprocal Rank (MRR). K values for hit ratio: 1, 5, 10, 20.
- Pre-Made Data Splits: The data is already split based on users for testing hypotheses, and the system must adhere to this split.
- Data: The system will work with MOOC course data, which is already preprocessed to some extent.
- Model Selection & Hyperparameter Tuning: Include model selection and tuning while considering the need for efficiency due to limited computational power.
Additional Notes:
* My data is interaction-only, meaning there are no user demographics or course content features. The recommendation system should focus solely on historical interaction data.
* I will provide either the data files (train and test) or the code used to create the splits.
Ideal Skills and Experience:
- Experience with Transformers4Rec and recommendation systems.
- Strong proficiency in Python and relevant machine learning libraries (e.g., PyTorch, NVTabular).
- Ability to work with custom data splits, with attention to maintaining the integrity of these splits for hypothesis testing.
- Prior experience with academic coding projects is a plus.
The final deliverable should be a fully functional system, seamlessly integrated with my data, ready for multiple rounds of training and testing. Full code is enough, I will do the rest.
I am looking for a complete implementation of a Transformers4Rec-based sequential next-course recommendation system. The system needs to work with my pre-made data splits to ensure rigorous hypothesis testing, and the splits are based on full user sequences. The project should include model selection and parameter tuning, considering my limited computational resources and the need to retrain the model multiple times.
Key Requirements:
- End-to-End Solution: From NVTabular schema creation to model training and evaluation.
- Metrics for Evaluation: Hit ratio at k, NDCG@k, Coverage (percentage of courses recommended), Mean - Reciprocal Rank (MRR). K values for hit ratio: 1, 5, 10, 20.
- Pre-Made Data Splits: The data is already split based on users for testing hypotheses, and the system must adhere to this split.
- Data: The system will work with MOOC course data, which is already preprocessed to some extent.
- Model Selection & Hyperparameter Tuning: Include model selection and tuning while considering the need for efficiency due to limited computational power.
Additional Notes:
* My data is interaction-only, meaning there are no user demographics or course content features. The recommendation system should focus solely on historical interaction data.
* I will provide either the data files (train and test) or the code used to create the splits.
Ideal Skills and Experience:
- Experience with Transformers4Rec and recommendation systems.
- Strong proficiency in Python and relevant machine learning libraries (e.g., PyTorch, NVTabular).
- Ability to work with custom data splits, with attention to maintaining the integrity of these splits for hypothesis testing.
- Prior experience with academic coding projects is a plus.
The final deliverable should be a fully functional system, seamlessly integrated with my data, ready for multiple rounds of training and testing. Full code is enough, I will do the rest.
Related categories:
Python
Machine Learning (ML)
Deep Learning
Recommendation System
Transformer Model