Semester Project Ideas for Recommender Systems - Please read the description

AI Discussions
,
1

Recommender Systems Course Project

Good-day everyone. I am an undergrad student and am currently taking a course on Recommender Systems. The course requires a mandatory project that can fall into one of the following four categories:

1. Innovative Improvement

  • Propose an improvement to an existing method.

  • Select a method from a top-tier conference paper (2023 or newer).

  • Implement both the original method and your proposed improvement.

  • Conduct experiments using all datasets from the original paper and compare the results.

  • The complexity of the original method may vary, and the improvement can be intricate or simple.

  • The scope of work, workload, and team size should be justified accordingly.

2. Comparative Analysis of Three or more Methods

  • Critically and empirically evaluate three or more well-known methods that solve the same problem.

  • Implement all three methods and perform a comparative experimental analysis.

  • Highlight key differences, strengths, and limitations of each method.

  • Provide well-supported observations and insights based on your findings.

3. Broad Comparative Evaluation

  • Evaluate five or more established methods for a given problem.

  • Implement these methods and demonstrate results across multiple datasets.

  • A thorough comparative analysis of performance will be appreciated.

4. Term Paper

  • Select a topic and write a comprehensive term paper by reviewing and analyzing all major published results in the field.

I want to work on the 1st and specifically on Graph based models but I don’t know how to start. As for the knowlege to work on Graph ML I’ve been intrested in them so I learnt about them. I would like to someone to point me right direction and as the deadline is on 25th Feb I would like to finalize the project idea so that I can work on it.

Thank you for your time to read this and I would like to request to help me in this.

Thank you everyone.

2

Hi Vivek,

I often use Chatgpt as a Virtual Secretary that is available to me 24/7 AND does not require a monthly salary to be paid at the end of the month.

In that point of view - the first answer that came to my mind when I saw your question is - I should tell this person to regularly use ChatGPT instead of Google in order to save time and enhance productivity.

Now, to answer your question - I asked chatgpt about your requirement.

Please use if you find it useful.

1. Personalized Graph-based Recommender with Temporal Attention

:pushpin: Idea:
Most Graph Neural Network (GNN)-based recommenders do not adequately model how user preferences evolve over time. This project can introduce a temporal attention mechanism that dynamically adjusts recommendation weights based on recent interactions.

:hammer_and_wrench: Steps to implement:

  • Select an existing GNN-based recommendation model (e.g., LightGCN, GAT-based recommender).
  • Introduce a temporal attention mechanism that gives more weight to recent interactions.
  • Train and test on standard datasets (MovieLens, Amazon Reviews, etc.).
  • Compare with the baseline model to measure improvements.

:mag: Innovation: Improved recommendations by incorporating time-sensitive user behavior.

2. Graph Contrastive Learning for Cold Start Recommendations

:pushpin: Idea:
Cold-start users (new users with few interactions) present a challenge for recommender systems. This project can use Graph Contrastive Learning (GCL) to enhance representations for such users by leveraging user-item graph structures.

:hammer_and_wrench: Steps to implement:

  • Start with an existing GNN-based recommender system (e.g., NGCF, GraphRec).
  • Apply contrastive learning techniques to generate better embeddings for cold-start users.
  • Experiment with different augmentation strategies (e.g., node dropping, edge masking).
  • Compare results with traditional collaborative filtering approaches.

:mag: Innovation: Better recommendations for new users with limited interactions.

3. Fairness-aware Graph Neural Networks for Recommender Systems

:pushpin: Idea:
Many recommender systems have inherent biases (e.g., popularity bias, demographic bias). This project can introduce fairness constraints into a GNN-based recommender system to ensure balanced recommendations across different user groups.

:hammer_and_wrench: Steps to implement:

  • Choose an existing GNN-based recommender (e.g., LightGCN, PinSage).
  • Integrate a fairness-aware loss function (e.g., equal representation across different demographic groups).
  • Experiment with real-world datasets that contain demographic attributes (e.g., Last.fm dataset).
  • Compare fairness-aware recommendations with standard recommendations.

:mag: Innovation: Reducing bias in graph-based recommender systems.

4. Hypergraph-based Recommender for Multi-Interest Users

:pushpin: Idea:
Most GNN-based recommenders use simple user-item bipartite graphs, but they fail to capture users’ multiple interests effectively. This project can use Hypergraphs, where nodes represent multiple user interests simultaneously.

:hammer_and_wrench: Steps to implement:

  • Extend a baseline GNN recommender with hypergraph neural networks.
  • Define hyperedges that group users/items with similar multi-interest profiles.
  • Train the model on diverse datasets (MovieLens, Goodreads, etc.).
  • Compare performance with traditional bipartite graph models.

:mag: Innovation: More accurate recommendations for users with diverse preferences.

Thanks,

Vishnu

1 Like