You are on page 1of 9

Joint Clustering and Ranking from

Heterogeneous Pairwise Comparisons


Chen-Hao Hsiao and I-Hsiang Wang

Graduate Institute of Communication Engineering,


National Taiwan University, Taipei, Taiwan
Email: {f07942062,ihwang}@ntu.edu.tw
ISIT 2021
1
Goal: Joint Clustering and Ranking

Rock
POP Goal:
1.Recover the underlying clusters
2.Rank the items in each cluster
Given only pairwise preferences!
Utilize the heterogeneity!

EDM Intra-cluster comparisons


Inter-cluster comparisons
2
Fundamental Questions
To achieve reliable joint clustering and ranking:
• How high of the heterogeneity do we need?
• How many comparisons do we need?
• How to utilize the heterogeneity?

3
Main Contribution

1. Characterize the fundamental limit for exact joint clustering and ranking
𝐥𝐨𝐠 𝒏
Level of heterogeneity × Number of comparisons = 𝚯( )
𝒏

2. Develop a polynomial-time three-step algorithm

4
Main Result

2
log 𝑛
𝑝𝐿 × Δ = Θ( )
𝑛
Expected number of comparisons Level of Heterogeneity

Impossible Achievable
By Fano’s inequality By the proposed algorithm
𝑝𝐿Δ2 𝐶1
log 𝑛
𝐶2
log 𝑛
𝑛 𝑛

5
Proposed Three-step Algorithm

Ranking Transformation Clustering

6
Proposed Three-step Algorithm

Borda
Elimination SDP
Counting

Ensures the
Transform the graph
correct ordering Solve the clustering task
into a block model
in each cluster
with high probability
7
Conclusion
• Information Limit:

Impossible Achievable
By Fano’s inequality With a polynomial-time algorithm
𝑝𝐿Δ2 𝐶1
log 𝑛
𝐶2
log 𝑛
𝑛 𝑛
• Proposed algorithm:
Borda
Elimination SDP
Counting
(Transformation) (Clustering)
(Ranking)
8
Thank you for listening!

You might also like