Hedonic Game Related Research

Title:Welfare maximization in fractional hedonic games

authors:Haris Aziz, Serge Gaspers Joachim, Gudmundsson, Juli´an Mestre

• hedonic game: comprises a set of agents who express preferences over coalitions they are they are present in and outcomes are partitions of the agents into disjoint coalitions. It provides a natural framework to study coalition formation.
• fractional hedonic games: each vertex of the network can be considered as an agent. An agent $i's$ valuation $v_i(j)$ of an agent $j$ can be represented by the weight of the directed edge $(i,j)$. Agent $i's$ valuation of a coalition $S$ of agents is the mean valuation $\sum_{j\in S} v_i(j)/\mid S \mid$ of the members of $S$.

A hedonic game $(N,\succeq)$ is said to be a fractional hedonic game(FHG) if for each player $i$ in $N$ there is a value function $v_i$ such that for all coalitions $S,T\in N$,$S\succeq_i T$ if and only if $v_i(S)\ge v_i(T)$.

Main contributions:

1. simple examples that show that utilitarian, egalitarian, and Nash welfare maximizing outcomes need not coincide, even in simple symmetric FHGs.
2. a reduction that shows that maximizing utilitarian welfare, egalitarian welfare, or Nash welfare is NP-hard, even for simple symmetric FHGs.
3. a ploynomial-time 2-approximation algorithm for maximizing the utilitarian welfare of simple symmetric FHGs.
4. a polynomialtime 4-approximation algorithm for maximizing the utilitarian welfare of symmetric FHGs.
5. a polynomial-time 3-approximation algorithm for maximizing the egalitarian welfare of simple symmetric FHGs

Related work

• hedonic games based on graphs examined from a social welfare perspective.
• a related class of hedonic game: additive seperabel hedonic games, consider welfare maximizing or stable partitions of FHGs and why stable or efficient outcomes of FHGs provide better clusterings.
• Olsen[2012] examined a variant of FHGs and considered computation of Nash stable outcomes.
• the prior work on FHGs, most of the focus has been on stabel partitions, this has a disadvantage that a stable outcome may not be guaranteed to exist[Aziz et al., 2014; Bilo` et al., 2014; Brandl et al., 2015] or may suggest the partition consisting of the grand coalition [Bilo` et al., 2014].

Theorem

Definitions

• symmetric: an FHG is said to be symmetric if $v_i(j) = v_j(i)$
• simple: an FHG is said to be simple if $v_i(j)\in \{0,1\}$
• notions of welfare of a partition $\mu$ of $N$
  
  1. utilitarian welfare: (sum of utilities) $$\sum_{i \in N} v_i(\mu(i))$$
  2. egalitarian welfare: (utility of worst off agent) $$min_{i \in N}v_i(\mu(i))$$
  3. Nash welfare: (product of utilities) $$\prod_{i\in N}v_i(\mu(i))$$

Theorem:

1. For simple symmetric FHGs, utilitarian welfare and egalitarian welfare are both NP-hard.
2. For simple symmtric FHGs, Nash welfare is NP-hard.
3. For simple symmtric FHGs, utilitarian welfare has a linear-time 4-approximation algorithm.
4. For simple symmtric FHGs, utilitarian welfare has an $O(\sqrt{|N|}|E|)$ time 2-approximation algorithm.
5. For symmtric FHGs, utilitarian welfare has a polynomial-time 4-approximation algrithm.
6. For simple symmtric FHGs, egalitarian welfare has a polynominal-time 3-approximation algorithm.

Hedonic Game Theorem

Definitions:

• finite set N of players
• coalition = non-empty subset of N
• partition $\Pi$ divides N into disjoint coalitions
• $\Pi (i)$ denotes coalition in $\Pi$ caontaining player $i \in N$
  
• every player $i \in N$ ranks all the coalitions containing i via $\preceq_i$ and $\prec_i$ ( that means player $i$ express its preferences through $\prec_i or \preceq_i$)
• a coalition S blocks a partition $\Pi$, if all players $i \in S$ have $\Pi(i) \prec S$ and hence strictly prefer being in S to being in current coalition $\Pi(i)$

Central definition:
A partition $\Pi$ is core stable, if there is no blocking coalition S.
Closely related:
weakly blocking coalition( no player worse; at least one player better off)
strongly core stable partition( no weakly blocking cailition)

Problem:
Given: set N with all the preferences of the players.
Decide whether there exists some core stable partition.

Main problem: $\exists\Pi\quad\forall S : \lnot (S\quad blocks \quad \Pi)$

Compansion problem: Given game andpartition, decide whether there is a blocking coalition.
$\Longleftrightarrow\quad \exists S:(S\quad blocks\quad \Pi)$
$\Longleftrightarrow$ Negation of inner problem $\forall S:\lnot(S\quad blocks\quad \Pi))$

Observation:
If the companion problen is solveable in polynomial time, then the main paoblem is contained in NP.

The Ballester encoding
…

Preferences from graphs
friend-oriented
[Dimitrov, Borm, Hendrickx, Sung,2006] propose preference structures based on directed graphs G = (N,A).
An arch$(x,y)$ from player $x$ to $y$ means that $x$ considers $y$ as a friend. Set $F_x$ contains the friends of player $x$, and set $E_x$ contains his enemies.
Definitions
Player $x$ prefers S to T (“$S \prec_x T$”), if and only if

1. $|S \cap F_x|>|T \cap F_x|$ or
2. $|S \cap F_x|=|T \cap F_x|$ and $|S \cap E_x|\le|T \cap E_x|$

Theorem: under friend-oriented preferences, there always is a core stable partition.(Proof:11/30)

enemy-oriented
[Dimitrov & al 2006] also discuss graph-based enemy-oriented preferences.
Definitions
Player $x$ prefers S to T (“$S \prec_x T$”), if and only if

1. $|S \cap E_x|<|T \cap E_x|$ or
2. $|S \cap E_x|=|T \cap E_x|$ and $|S \cap F_x|\ge|T \cap F_x|$

If player $x$ does not like player $y$, then in any core partition $x$ and $y$ must not in the same coalition.
Assume that friendship is mutual/symmetric, and use undirected graphs. In core stable partition, every coalition is a clique.

Theorem

• Under enemy-oriented preferences, there always is a core stable partition.
• Under enemy-oriented preferences, compansion-probelm is NP-complete.(Proof:13/30)

Additive preferences
Every palyer$x$ assigns to every other player $y$ a real value $v_x(y)$. High positive value = very attracted; negative value = aversion.

Player $x$ values coalition $S$ as $v_x(S) = \sum_{s \in S}{v_x(s)}$
Player $x$ ranks $S \preceq_x T$ if and only if $v_x(S) \le v_x(T)$

Theorem
Under additive preferences, the compansion-problem is NP-complete.[Sung, Dimitrov, 2007]
Under additive preferences, the main-problem is $\sum_2^p$-complete. [Woeginger, 2012]