A Game-theoretic Analysis of Catalog Optimization JOEL OREN, UNIVERSITY OF TORONTO. JOINT WORK WITH: NINA NARODYTSKA, AND CRAIG BOUTILIER 1 Motivating Story: Competitive Adjustment of Offerings •A large retail chain opens a new store. •Multiple competitors. •Multiple potential customers: • Typically doesn’t buy too many items – say, just one item. • Buy their most preferred item, given what is offered in total – over all stores. •Exogenous (fixed) prices. •How should they choose what to offer, so as to maximize their profits? •A form of assortment optimization. 2 • Catalog: a set (assortment) of offered items. • Best-response: Optimizing one’s catalog may be tricky – ≻ 1 50 ≻ ⋯ Vendor 1 $10 ≻ ≻ 50 $5 $4 $15 $8 Catalog 1 $10 1 Vendor 2 Catalog 1 $15 $8 ≻ ⋯ X 100 • What are the convergence properties of these dynamics? 1. Do pure Nash eq. (PNE) exist? 2. What is the Price of Anarchy/Stability, (PoA/PoS)? $10 ≻ ≻ ≻ ≻ 3 The Formal Model •The Catalog Selection game: 𝐺 = 𝐶 1 , … , 𝐶 𝑘 , 𝒑, 𝑁, 𝝅 . •𝑘 strategic vendors (agents): • Sets of items (𝐶 1 , … , 𝐶 𝑘 ) (not necessarily disjoint). Total # of items is 𝑚 = 𝑖 𝐶 . Each item has unlimited number of copies. 𝑖 • An exogenous price vector 𝑝 ∈ 𝑅+𝑚 . • Vendor 𝑖’s strategy: a catalog 𝑅 ⊆ 𝐶 𝑖 . Strategy profile 𝐒 = (𝑆1 , … , 𝑆 𝑘 ). • Each vendor’s goal is to maximize revenue, 𝑟𝑖 (𝑆𝑖 , 𝑺−𝒊 ). •Set N of unit-demand consumers with rankings 𝜋 = (𝜋1 , … , 𝜋𝑛 ) over • Each consumer 𝑖 buys her most preferred item 𝑐𝑖 in 𝑖. 𝑆 𝑖 𝑖. 𝐶 𝑖 4 Selecting the Best Response – the Full Information Setting •The Full Information setting: the consumers’ preference profile 𝝅 is commonly known. •Given 𝑺−𝒊 , how should vendor 𝑖 select 𝑆 𝑖 ? ◦ Cheap items in 𝐶 𝑖 may be commonly preferred over expensive items in 𝐶 𝑖 . ◦ Not adding certain items in 𝐶 𝑖 -- may lose consumers due to competition. Theorem: Computing a best response is Max-SNP hard. Implication: there is a constant, such that approximating the maximal profit beyond this constant is NP-hard. 5 A Special Case: Single-Peaked Truncated Preference Single-peaked, L-truncated preferences: if looking at the prefixes composed of only vendor 𝑖: they’re of length 𝐿, and they are single-peaked. •Result: we can optimize vendor 𝑖’s best-response using a dynamicprogramming approach. •See paper for details. 6 Partial Information Setting •Consumer rankings are unknown in advance. Instead, they are drawn from a commonly known distribution 𝐷. •Best response: 𝑎𝑟𝑔𝑚𝑎𝑥𝑆⊆𝑅𝑖 𝐸𝝅∼𝐷 [𝑟𝑖 𝑆, 𝑹−𝒋 ] . •Warmup: preferences are drawn u.i.d. from the complete set of preferences. • Idea: greedily add items until expected revenue starts to decline. Vendor 1 $10 Vendor 2 $5 $4 $15 Catalog 1 $10 ≻ Catalog 1 $1 5 ≻ $8 $8 ≻ ≻ 𝐷 7 Partial Information Setting •Consumer preferences are unknown in advance. Instead, they are drawn from a commonly known distribution 𝐷. •Best response: 𝑎𝑟𝑔𝑚𝑎𝑥𝑆⊆𝑅𝑖 𝐸𝝅∼𝐷 [𝑟𝑖 𝑆, 𝑹−𝒋 ]. •Mallows 𝜑 −distribution: each 𝜋𝑖 ∼ 𝐷(𝜋, 𝜑), where Pr 𝜋 = 𝜑 𝑑𝐾 (𝜋,𝜋) /𝑍𝑚 , and 𝑑𝐾 (⋅,⋅) is the Kendall’s 𝜏distance. • Result: There is a polytime DP algorithm for optimizing the best response. • Algorithm can be generalized to handle mixtures of Mallows distributions; i.e., lotteries [ 𝑝1 , 𝐷1 , 𝑝2 , 𝐷2 , … , 𝑝𝑟 , 𝐷𝑟 ]. Vendor 2 Vendor 1 $10 $4 $5 $8 $15 Catalog 1 Catalog 1 $15 $10 ≻ $8 ≻ ≻ ≻ 𝐷 8 Equilibria and Stability of the Game •The Catalog Selection game: 𝐺 = 𝐶 1 , … , 𝐶 𝑘 , 𝒑, 𝑁, 𝝅 All vendors’ Mutual sets are sets disjoint •Does this game admit PNE? If so, what are the guarantees on them? •The social outcome: total profit. Full information •Price of Anarchy (PoA): ratio of the OPT social Partial to the worst-case PNE. •Price of Stability (PoS): ratio of the OPT social to the best PNE. information - IC 9 Full Information, Disjoint Sets •𝝅 = (𝜋1 , … , 𝜋𝑛 ), 𝐶 𝑖 ∩ 𝐶𝑗 = ∅ for all 𝑖 ≠ 𝑗. •There exist instances of the game w/o PNE. 𝑥+𝜖 ≻ 𝑥+𝜖 ≻ 2𝑥 ≻ 2𝑥 𝑥+𝜖 𝑥+𝜖 𝑥+𝜖 ≻ ≻ 2𝑥 2𝑥 ≻ ≻ ≻ 𝑥+𝜖 2𝑥 ≻ 2𝑥 𝑥+𝜖 (2𝑥 + 2𝜖, 𝑥 + 𝜖) (2𝑥 + 2𝜖, 2𝑥) (2𝑥, 2𝑥 + 2𝜖) (4𝑥, 2𝑥 + 𝜖) 2𝑥 𝑥+𝜖 2𝑥 10 Partial Information, Disjoint Sets •𝜋𝑖 ∼ 𝐷 𝜋, 𝜑 . If 𝜑 = 0, all preferences equal 𝜋 -- trivial PNE. So assume 𝜑 = 1 – an Impartial Culture (IC) – preferences are drawn u.i.d. •Result: there always exists a PNE. • Intuition: 1. Given 𝑹−𝒋 , best set of size 𝑡 is the set of 𝑡 most expensive items in 𝐶𝑗 . 2. If number of other vendors’ item increased – vendor 𝑗 can only bestrespond by adding items, never removing items. •Result: The POS is Ω(𝑚) – the social welfare of the best PNE can be a 1 −fraction of the optimal welfare. 𝑚 11 Full Information, Mutual Sets •The preference profile 𝝅, is fully known. 𝐶 1 = ⋯ = 𝐶 𝑘 . •If an item is offered by 𝑟 vendors, payment for it is split among them evenly. •Observation: there is always a PNE: (𝐶 1 , … , 𝐶 𝑘 ) – no vendor has an incentive to remove any items. •Result: The PoS is Ω(2𝑚 ). • Use a price vector that constitutes an “approximately” geometric series: 1 4 1 (1, 2 + 𝜖, + 2𝜖, … ). A single consumer who ranks items in increasing order of price. • Cheapest item will always be offered, and bought. 12 Partial Information, Mutual Sets •Preferences uphold the Impartial Culture assumption. •A PNE always exists. •PoA: at least Ω(𝑚). Idea: one item of value 1, 𝑚 − 1 items of value 𝜖. If 1 everyone selects all items, first item is picked with probability . 𝑚 𝑚⋅𝑘 𝑂( ) log 𝑚 •Price of Stability: – ratio of the optimal pure Nash eq. to the optimal social welfare. • Idea: constructively find a Nash eq., by adding items in decreasing order prices; lower-bound value upon termination. 13 Conclusions and Future Directions •Best response: Max-SNP hard in general, easier under some assumptions. • Question: can we design an approximation algorithm for the full-info. case? •Game theoretic analysis: PoA/PoS under complete VS partial information, disjoint VS mutual sets. • Question: can we show that there is always a PNE under a general Mallows dist.? •Additional directions: • Other classes of preferences. • Study the game when prices are set endogenously. 14
© Copyright 2025 Paperzz