Hidden Markov Models and Viterbi Decoding: Inferring a Customer's Hidden Decision Stage from Click Sequences and Timing the Offer: chapter answers
Online shops often judge how far a visitor has come from her last click and decide on that basis whether to add a quote. This book refutes that common sense. A customer's decision stage is invisible and each click is only a blurred projection of it; put the whole click sequence into a hidden Markov model and infer the current stage with the forward filter, and the judgement is more accurate, by an amount that can be computed. After checking Rabiner's tutorial of 1989, the book corrects its project blueprint: the tutorial gives algorithms and two optimality criteria but measures no accuracy of…
DOI pending · monograph page goes live after publication · 5DL-B-300 · 中文版
- Chapter 1 · The Last-Click Illusion: Whole Sequences against a Single Page
Judging a customer's stage from the last click alone gives, with the illustrative parameters, a four-way error rate of 0.5180 and a ready-or-not accuracy of 0.6990. The forward filter over the whole sequence raises four… - Chapter 2 · The Viterbi Trellis: Dynamic Programming in Logarithms, Backtracking and Underflow
Chapter 2 unfolds Viterbi on a trellis, shows direct multiplication underflowing near click 368, and finds that the best path holds only 0.0357 of the total probability. - Chapter 3 · Forward and Backward: Posterior Marginals, the Best Path and "Wait Two More Clicks"
Chapter 3 covers forward-backward and the two optimality criteria; judging two clicks later adds 0.0413 in accuracy. - Chapter 4 · Baum–Welch: Estimating Parameters from Unlabelled Clickstreams
Chapter 4 estimates parameters from unlabelled clickstreams with Baum–Welch, and ten random starts decode with accuracy from 0.3877 to 0.5181. - Chapter 5 · Identifiability and Label Switching: What Clickstreams Cannot Say
Chapter 5 treats label switching, the number of stages and weak identifiability: naming stages by average position within a visit gives an accuracy of 0.4477, close to 0.4623 with true labels, and with blurred pages cli… - Chapter 6 · Timing the Offer: the Inquiry as Evidence, a Cost Threshold and Calibration
Chapter 6 gives the timing rule: never speak first, always answer price questions, one price for all, urgency only from reality, and add a quote to other answers only by a calibrated posterior and a cost threshold; coun… - Chapter 7 · Classical Sources East and West: Markov's Letter Chain in Onegin and the Chuai Chapter of the Guiguzi
Chapter 7 examines Markov's letter chain in Onegin of 1913 (stationary vowel share 0.4319) and the Chuai chapter of the Guiguzi (one click carries only 0.2680 bits). - Chapter 8 · Closing the Loop: a Weekly Stage Review, the Customer Stage Decoder and Pre-Registered Tests
Chapter 8 gives the weekly review, the Customer Stage Decoder and pre-registered tests: test one needs 372 visits with a self-reported stage, test two 5301 chats per arm. Marked boxes separate published evidence, runnab…
Illustrative model calculation, not a live quote