Notes

LightFM
1. works well in cold-start scenarios
2. Matrix factorisation better than neighbourhood method
3. used linear combinations
4. user and items are related as latent vectors
5. LightFM performs well in cold-start and low density scenarios.
6. Warm-start, dense user-item matrix, lightFm performs at least as well as the MF model

References:
[1] https://www.cs.utexas.edu/~ml/papers/cbcf-aaai-02.pdf
Content-Boosted Collaborative Filtering for Improved Recommendations

Description:
This paper proposes a new approach called Content-Boosted Collaborative Filtering in building hybrid recommendation system. They (P. Melville, R.J. Mooney, R. Nagarajan) improve existing user data using content based filtering then use Collaborative Filtering to improve suggestions.

They create a pseudo user-ratings vector for every user and combine all the vectors to create a dense pseudo user-rating matrix. The matrix is then used to perform collaborative filtering. Using Pearson Correlation, the similarity between the current user and another user is calculated. But, the problem with this approach is that the accuracy is dependent on the number of movies a user has rated. If the user has not rated many items, the pseudo user-rating vector will be inaccurate thus it won't recommend the most related movies to the current user.

However, lightFM performs well in cold-start and less dense situations as it has latent representation of user and item features.

[2] Ya-Yueh Shih, Duen-Ren Liu
Hybrid recommendation approaches: collaborative filtering via valuable content information.
https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=1385682
Proceedings of the 38th Hawaii International Conference on System Sciences - 2005

A research conducted by Ya-Yueh Shih and Duen-Ren Liu included Hybrid recommendation approaches as described in [8]. They used collaborative filtering via valuable content information [8]. This hybrid filtering uses two different approaches, the first approach is Collaborative filtering and the second approach is the customer demands regarding an item. They combined both to enhance the prediction accuracy for the recommendation system.

The problem with this approach is that if the number of users are limited, it will result in a low-quality recommendation. This approach is dependent on data from customers including what their choices and demands were. My technique will provide a satisfactory match even if the customers demand is low or there are less customers available.

[13] Vozalis, M., & Margaritis, K. G. (2004). On the combination of user-based and item-based collaborative filtering. International Journal of Computer Mathematics, 81(9), 1077–1096.

[Reference 13] has a different approach. They (Manolis Vozalis and Konstantinos G. Margaritis) combined user-based and item-based Collaborative Filtering scheme. They have two different algorithms named Hybrid-lb and Hybrid-CF. Hybrid-lb analyzes a big area of related users and then uses this batch to develop the item-based recommendation model. Hybrid-CF starts by finding an item, which matches the one we want a prediction for, and based on that batch, it generates its user-based predictions. [13]

Hybrid-lb needs a huge amount of users input and it won't perform well in cold-start situations.

https://www.tandfonline.com/doi/full/10.1080/03057920412331272199?scroll=top&needAccess=true

[14] ] Kumar, N. P., & Fan, Z. (2015). Hybrid user-item based collaborative filtering. In Procedia Computer Science (Vol. 60, pp. 1453–1461). Elsevier B.V.

Nitin Pradeep Kumar and Zhenzhen Fan have a different approach to build a hybrid recommendation model on their research Hybrid user-item based collaborative filtering as describe in [14]. Since Collaborative Filtering (CF) suffers from flexibility and data shortage, they mixed CF with CBR (Case-Based Reasoning) and SOM (Self-Organizing Map) enhanced with GA (Genetic Algorithm) to make a better prediction.

[15] Burke, Knowledge-based recommender systems. In A. Kent (ed.), Vol. 69, Supplement 32. New York: Marcel Dekker, 180-200 2000.

Hybrid recommendation systems are the integrative, parallel, or linear combinations of several recommendation systems with an effort to fill in the gaps of single recommendation systems. Top-N based collaborative filtering (TNCF) and majorizing similarity based collaborative filtering (MSCF) [18] proposed by Song are hybrid collaborative filtering approaches which integrate score similarity and property similarity. They first compute user similarity and select the top N nearest neighbors of the target user and then predict scores and provide recommendation. This method improves the accuracy, while it greatly increases the complexity of the computation.

Collaborative filtering recommendation, content-based recommendation, and knowledge-based recommendation approaches are all based on a single information source and fail to satisfy users’ diversified demand and effectively solve the cold start and data sparsity problems.
Although hybrid recommendation approaches try to overcome the cold start and data sparsity problems by combing several recommendation systems, they are just linear combinations and cause high approach complexity and non-accurate prediction.
In addition, these approaches proposed to solve data sparsity fail to consider the effects of users’ influences and prediction order on recommendation accuracy.