# Identifying Power User in MF-Based Recommendation System
Power users are those who can exert considerable influence over the recommendations presented to other users.
A gap in the research is that characteristics of influential/power users have largely been ignored. How are power users best identified and selected in a RS? What constitutes an influential user and how specifically such users, individually or in groups, impact other users in the system.
Much of the previous work in this area does not extend beyond collaborative recommendation based on the standard k- Nearest-Neighbor (kNN) approach to more recent approaches such as matrix factorization.
## Prior Work: Measures of Influence
### Influences in social network
- "Centrality": degree centrality & distance centrality
- Linear Threshold Models, Independent Cascade Models?
- Mean taste similarity & dispersion
### Algorithm dependent measures
- CF algorithms:
- Expected Lift in Profit
- Similarity Links
- Influence Discrimination Model (k-NN & NMF)
### Algorithm independent measures
- Number of Prediction-Differences

**Ref:**
1. [Influence in Ratings-Based Recommender Systems: An Algorithm-Independent Approach](http://files.grouplens.org/papers/RashidAl_siam05.pdf "title")
2. [Power of the Few: Analyzing the Impact of Influential Users in Collaborative Recommender Systems](https://arxiv.org/pdf/1905.08031.pdf "title")
3. [The Structure of Social Influence in Recommender Networks](https://pure.mpg.de/rest/items/item_3195270/component/file_3231036/content "title")
## Related Work
- Shilling attack/profile injection attacks: Attacks on RSs by providing false ratings.
- Power User attack: A set of power user profiles with biased ratings influence the results presented to other users.