Proof of Lemma 1, p. 53 — with no prior click the next click is Bernoulli(pᵢ): the hook probability equals the coupled first-click probability and 1 − ∏(1 − p)
ProvedRankAssort.SimpleLearn.first_click_identityFix a customer type and a ranking (more generally any duplicate-free list of products), and let be the browsed window. Then the probability that the customer is hooked under the sequential click law with bias equals
- the probability, under the coupled first-click law (one Bernoulli like-variable per product), that some product in the window is liked:
- the closed form
This is the content of the paper's remark that, conditional on no prior clicks, a click on product is a Bernoulli variable with probability . It ties the coupled first-click law used in the proofs to the model: every statement phrased with the coupled law is a statement about the sequential click law with bias.
Formalization Note Ranks are 0-based in Lean: the paper's rank is index , a ranking is a duplicate-free list of the products (Fin n), and is the list entry at index . The first part is hookProb θ l = coupledProb θ (…), the second the product formula over the window Fin (window θ l). The hypothesis that the list has no duplicates is needed: the coupled law gives every copy of a product the same coin.
Differences from the source:
- generalization: any ranking π → any duplicate-free list l of products (a ranking or a prefix of one); reason: a full ranking is the case length l = n
- reading: conditional on no prior clicks, the click on product i is Bernoulli(pᵢ + fᵢ(∅)) = Bernoulli(pᵢ) → the conditional hook probability equals the coupled first-click probability that some window product is liked, and equals 1 − ∏ over the window of (1 − p); reason: the remark is used in the proof of Lemma 1 exactly to replace the sequential law by these forms; the per-click statement is their one-step content
- encoding: conditional on the customer's type (p, f, k) → a statement for a fixed θ : CustomerType n; reason: the same conditioning
- encoding: ranks 1..n, a ranking is a permutation π of [n], π⁻¹(j) the product at rank j → 0-based indices, a ranking is a duplicate-free List (Fin n) of length n, π⁻¹(j) is the entry at index j − 1; reason: Lean convention; the same rankings
import Mathlib import Definitions.Def_RankAssort_SimpleLearn_Model
namespace RankAssort.SimpleLearn
theorem first_click_identity {n : ℕ} (θ : CustomerType n) (l : List (Fin n)) (hl : l.Nodup) :
hookProb θ l =
coupledProb θ (fun L => ∃ j, ∃ hj : j < l.length, j < θ.k ∧ L (l.get ⟨j, hj⟩) = true) ∧
hookProb θ l = 1 - ∏ j : Fin (window θ l), (1 - θ.p (windowGet θ l j)) := by sorry
end RankAssort.SimpleLearn
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.