Engineering Intentional Bias: The MIkCDSP Approach to Smarter Online Surveys

Executive Summary

TL;DR: Researchers have developed a new graph-theoretical framework called MIkCDSP to select survey respondents from social networks. Unlike traditional methods that strive for "unbiased" representative samples, this method allows researchers to deliberately amplify minority opinions (like product complaints) by limiting the "cohesiveness" of selected groups while ensuring they still "dominate" (represent) the entire network.

Positioning: This work bridges the gap between traditional social dominance theory and targeted marketing analytics, moving from "unbiased sampling" to "strategically biased representation."

The Paradox of Unbiased Surveys

In marketing or political science, we are taught that "bias is bad." However, the authors identify a critical exception: Product Quality Management. If 95% of users are happy with a smartphone, a perfectly representative sample will yield 95% "satisfaction" feedback. For a manager looking to fix bugs, that 95% is noise. They need the 5%—the frustrated minority.

The challenge is finding a subset of people who represent the whole population (Representation) but are not clustered in the majority "echo chamber" (Controlled Bias).

Methodology: The Inverse k-Core Dominating Set (IkCDS)

The authors propose that people with similar opinions cluster together in social networks. To capture the minority, we must pick a Dominating Set (DS)—a group where every person in the network is either in the group or friends with someone in the group—but we must restrict its internal connectivity.

1. Mathematical Innovation

They define the Inverse k-Core Dominating Set:

• Dominating Set Property: Ensures everyone's "voice" is reachable.
• Inverse k-Core Property: For every selected respondent $v$, the number of their neighbors also in the respondent group must be $\le k$.

By lowering $k$, you force the algorithm to pick "lonely" nodes or nodes from smaller clusters, effectively preventing the majority from dominating the sample.

2. The Algorithm: Greedy-MIkCDSA

Since the problem is NP-hard, the authors designed a greedy strategy. The algorithm iteratively selects nodes that cover the most "uncovered" people in the network while strictly adhering to the $k$-neighbor limit.

Figure 1: Mathematical formulation used to partition the graph and bound the greedy selection process.

Proof of Efficiency

The core of the paper is the rigorous proof of the approximation ratio. Using harmonic functions and vertex partitioning, the authors prove:

• Performance Ratio: $1+\ln \Delta$ (where $\Delta$ is the max degree).
• Feasibility: They prove a solution always exists, even for $k=0$ (which results in an Independent Set).

Figure 2: The final derivation showing the upper bound of the greedy solution compared to the optimal.

Critical Insight & Future Outlook

The beauty of this research lies in its Inductive Bias. It assumes social structure mirrors opinion structure. While highly effective for "professional" or "interest-based" networks, its performance on general-purpose networks (like a random Facebook graph) remains to be tested in the field.

Limitations:

• Requires knowledge of the network topology.
• The parameter $k$ is currently chosen heuristically; finding the "optimal $k$" for a specific desired bias level remains an open problem.

Future Work: The authors suggest extending this to weighted edges (where friendship strength varies) and multiple social networks to improve the robustness of the selected respondent group.