TDT4215: Recommender Systems (Group 7)

TASK: Develop a news recommender system

DATA: user data from the local digital newspaper (January - March)

• 1,000 most active users
• 9 attributes (userID, title)

Examples

• CF --> Explicit Matrix Factorization (MF)
• Content-based --> TF-IDF + Cosine similarity

EVALUATION

1. Recall (hit rate) --> positive instances that are correctly predicted
2. CTR (Click- through rate) --> recommendations clicked over total rec
3. ARHR (average reciprocal hit rate)
4. MSE

REPORT (max 15 pages)

1. Motivation
2. overview of existing news recommender algorithms and approaches
3. evaluation results
4. conclusion

EVALUATION: report + presentation + performance evaluation

STEPS

1. train test split --> scikit-learn
2. Implement a Collaborative filtering RS
3. Implement a Content-based RS
4. Merge them and build a Hybrid RS
5. Evaluation Recall, ARHR)

TODO for Content-based Filtering

Implement the article from https://medium.com/@bindhubalu/content-based-recommender-system-4db1b3de03e7 or from https://towardsdatascience.com/content-based-recommender-systems-28a1dbd858f5

1. Only work with single user file:

2. TF*IDF for keywords compared to the keywords of all their articles

3. Build matrix from the article

       doc0       doc1   doc2   doc3
   w0  tfidf00
   w1              ...
   w2                           tfidf23
   

4. Compute cosine similarity between items

5. Recommend from the article