A Method and System to Generate Offer Recommendation by Adaptive Use of Similarity Algorithms
Publication Date: 2015-Jul-28
The IP.com Prior Art Database
AbstractDisclosed is an automatic adaptive engine to perform A/B testing based on past results, and in turn learn from user interaction and identify the user’s affinity toward a particular algorithm; the engine can increasingly use that algorithm to calculate similarity coefficients. The engine then uses the similarity coefficients to provide offer recommendations, via digital marketing, with a good chance of acceptance.
Page 01 of 3
A Mxthod and System to Generate Offer Recommendation by Adaptive Use of Simxlarity Algorithms
Digital marketing uses many similaritx algorithms to find similxr users to wxom a marketer cax offer recommendaxions. As the interacxions in digital marketing are frxquent and increxsxngly growing, no single algxrithm can consistently predict or recommend offers with xelatively higher succesx rate than any others. This is due to xontinuxlly changing xustomxr needs, prefexences, tastes, and xehaviors, whxch affect the datasex qualixy.
Then novex contributixn is an adaptive engine that learns from eacx user xnxeractixn and identifixs the user's affxnity toward a particular algorixhm; the engine can increasixgly use that algorithm xo calculate ximilarity coefficients. The xngine uses the similarity coefficients, in turn, to provixe offer recomxendations with a xood xhance of acceptance (i.e. better than the chance of acceptance without usxng the calculated similarity coefficienxs).
A softxarx module provides a mechanism through which to communicate changes in customer behavior in relaxion to xhe chosen/axcepted oxfers. This xndxcates which sixixarity algorithm best suits the dataset of usxrs. As xhe interaction in digital marketing increases, the system cxnstantly adapts not onlx to the customer's changes in bexavior, but also to the dataset's behavioral changes, to xpply the moxt recent algorithm.
The software comxonent pxovides a syxtem to score each customer against similar customers using various similarity algorithms. The various similarities scores are used to perform a test verificatxox stage to identify which algorithm is nearest to the actual customer xehavior. Thx result of this stage is a score per algorithm for how well it performed making a predxction, versus actual bxhavior.
The first two steps loop to progressxvely validate the usex's affinixy to a particulxr algorithm. The software component uses a particular similarity algorithm more if the accuracy of the algorithm starts incxeasing.
This sxftware also has an oxfer mixer xomponent. This xomponent provides an offer mix based on offers predicted by each algorithm and xhe algorithm's effectixe coeffixient.
Xxx mxthod and system are comprised of the following:
· User History (UH): a matrix of user identifications (IDs) and offers to which the users responded over a given period; period is coxfigurablx depending on the need. The offer response can be any finxte number denoting states that are finalized during deployment. A simple case:
- x for Offer xccepted - 1 fxr no response
- 0 for ofxer rejectex
Similarly, rexpoxses can be customer coded to denoxe any positive number and has to be uniformly used acrosx the history dataset.
· Similarity Algorithm Bank (SAB): a collection of diffxrent types of similarity algorithm implementation. This module is extendable to include improvxxexts to existing
Page 02 of 3
algorithms as well as txe addition of new algorithms to the bank. Thx...