| Title: | Markov Model for Online Multi-Channel Attribution |
|---|---|
| Description: | Advertisers use a variety of online marketing channels to reach consumers and they want to know the degree each channel contributes to their marketing success. This is called online multi-channel attribution problem. This package contains a probabilistic algorithm for the attribution problem. The model uses a k-order Markov representation to identify structural correlations in the customer journey data. The package also contains three heuristic algorithms (first-touch, last-touch and linear-touch approach) for the same problem. The algorithms are implemented in C++. |
| Authors: | Davide Altomare [cre, aut], David Loris [aut] |
| Maintainer: | Davide Altomare <[email protected]> |
| License: | GPL-3 | file LICENSE |
| Version: | 2.0.7 |
| Built: | 2024-10-21 06:19:01 UTC |
| Source: | CRAN |
Advertisers use a variety of online marketing channels to reach consumers and they want to know the degree each channel contributes to their marketing success. This is called online multi-channel attribution problem. In many cases, advertisers approach this problem through some simple heuristics methods that do not take into account any customer interactions and often tend to underestimate the importance of small channels in marketing contribution. This package provides a function that approaches the attribution problem in a probabilistic way. It uses a k-order Markov representation to identify structural correlations in the customer journey data. This would allow advertisers to give a more reliable assessment of the marketing contribution of each channel. The approach basically follows the one presented in Eva Anderl, Ingo Becker, Florian v. Wangenheim, Jan H. Schumann (2014). Differently for them, we solved the estimation process using stochastic simulations. In this way it is also possible to take into account conversion values and their variability in the computation of the channel importance. The package also contains a function that estimates three heuristic models (first-touch, last-touch and linear-touch approach) for the same problem.
| Package: | ChannelAttribution |
| Type: | Package |
| Version: | 2.0.7 |
| Date: | 2023-05-17 |
| License: | GPL (>= 2) |
Package contains functions for channel attribution in web marketing.
Davide Altomare, David Loris
Maintainer Davide Altomare <[email protected]>
ChannelAttribution Official Website: https://channelattribution.io
Eva Anderl, Ingo Becker, Florian v. Wangenheim, Jan H. Schumann: Mapping the Customer Journey, 2014, doi:10.2139/ssrn.2343077
Estimate a Markov model from customer journey data after automatically choosing a suitable order. It requires paths that do not lead to conversion as input.
auto_markov_model(Data, var_path, var_conv, var_null, var_value=NULL, max_order=10, roc_npt=100, plot=FALSE, nsim_start=1e5, max_step=NULL, out_more=FALSE, sep=">", ncore=1, nfold=10, seed=0, conv_par=0.05, rate_step_sim=1.5, verbose=TRUE, flg_adv=TRUE)auto_markov_model(Data, var_path, var_conv, var_null, var_value=NULL, max_order=10, roc_npt=100, plot=FALSE, nsim_start=1e5, max_step=NULL, out_more=FALSE, sep=">", ncore=1, nfold=10, seed=0, conv_par=0.05, rate_step_sim=1.5, verbose=TRUE, flg_adv=TRUE)
Data |
data.frame containing customer journeys data. |
var_path |
column name containing paths. |
var_conv |
column name containing total conversions. |
var_null |
column name containing total paths that do not lead to conversions. |
var_value |
column name containing total conversion value. |
max_order |
maximum Markov Model order considered. |
roc_npt |
number of points used for approximating roc and auc. |
plot |
if TRUE, a plot with penalized auc with respect to order will be displayed. |
nsim_start |
minimum number of simulations used in computation. |
max_step |
maximum number of steps for a single simulated path. if NULL, it is the maximum number of steps found into Data. |
out_more |
if TRUE, transition probabilities between channels and removal effects will be shown. |
sep |
separator between the channels. |
ncore |
number of threads used in computation. |
nfold |
how many repetitions are used to verify if convergence is reached at each iteration. |
seed |
random seed. Giving this parameter the same value over different runs guarantees that results will not vary. |
conv_par |
convergence parameter for the algorithm. The estimation process ends when the percentage of variation of the results over different repetitions is less than convergence parameter. |
rate_step_sim |
number of simulations used at each iteration is equal to the number of simulations used at previous iteration multiplied by rate_step_sim. |
verbose |
if TRUE, additional information about process convergence will be shown. |
flg_adv |
if TRUE, ChannelAttribution Pro banner is printed. |
An object of class data.frame with the estimated number of conversions and the estimated conversion value attributed to each channel.
Davide Altomare ([email protected]).
## Not run: library(ChannelAttribution) data(PathData) auto_markov_model(Data, "path", "total_conversions", "total_null") ## End(Not run)## Not run: library(ChannelAttribution) data(PathData) auto_markov_model(Data, "path", "total_conversions", "total_null") ## End(Not run)
Find the minimum Markov Model order that gives a good representation of customers' behaviour for data considered. It requires paths that do not lead to conversion as input. Minimum order is found maximizing a penalized area under ROC curve.
choose_order(Data, var_path, var_conv, var_null, max_order=10, sep=">", ncore=1, roc_npt=100, plot=TRUE, flg_adv=TRUE)choose_order(Data, var_path, var_conv, var_null, max_order=10, sep=">", ncore=1, roc_npt=100, plot=TRUE, flg_adv=TRUE)
Data |
data.frame containing customer journeys. |
var_path |
column name of Data containing paths. |
var_conv |
column name of Data containing total conversions. |
var_null |
column name of Data containing total paths that do not lead to conversion. |
max_order |
maximum Markov Model order considered. |
sep |
separator between channels. |
ncore |
number of threads used in computation. |
roc_npt |
number of points used for approximating roc and auc. |
plot |
if TRUE, a plot with penalized auc with respect to order will be displayed. |
flg_adv |
if TRUE, ChannelAttribution Pro banner is printed. |
An object of class List with the estimated roc, auc and penalized auc.
Davide Altomare ([email protected]).
## Not run: library(ChannelAttribution) data(PathData) res=choose_order(Data, var_path="path", var_conv="total_conversions", var_null="total_null") #plot auc and penalized auc plot(res$auc$order,res$auc$auc,type="l",xlab="order",ylab="pauc",main="AUC") lines(res$auc$order,res$auc$pauc,col="red") legend("right", legend=c("auc","penalized auc"), col=c("black","red"),lty=1) ## End(Not run)## Not run: library(ChannelAttribution) data(PathData) res=choose_order(Data, var_path="path", var_conv="total_conversions", var_null="total_null") #plot auc and penalized auc plot(res$auc$order,res$auc$auc,type="l",xlab="order",ylab="pauc",main="AUC") lines(res$auc$order,res$auc$pauc,col="red") legend("right", legend=c("auc","penalized auc"), col=c("black","red"),lty=1) ## End(Not run)
Example dataset.
data(PathData)data(PathData)
Data is a data.frame with 10.000 rows and 4 columns: "path" containing customer paths, "total_conversions" containing total number of conversions, "total_conversion_value" containing total conversion value and "total_null" containing total number of paths that do not lead to conversion.
Estimate theree heuristic models (first-touch, last-touch and linear) from customer journey data.
heuristic_models(Data, var_path, var_conv, var_value=NULL, sep=">", flg_adv=TRUE)heuristic_models(Data, var_path, var_conv, var_value=NULL, sep=">", flg_adv=TRUE)
Data |
data.frame containing paths and conversions. |
var_path |
column name containing paths. |
var_conv |
column name containing total conversions. |
var_value |
column name containing total conversion value. |
sep |
separator between the channels. |
flg_adv |
if TRUE, ChannelAttribution Pro banner is printed. |
An object of class data.frame with the estimated number of conversions and the estimated conversion value attributed to each channel for each model.
Davide Altomare ([email protected]).
## Not run: library(ChannelAttribution) data(PathData) heuristic_models(Data,"path","total_conversions") heuristic_models(Data,"path","total_conversions",var_value="total_conversion_value") ## End(Not run)## Not run: library(ChannelAttribution) data(PathData) heuristic_models(Data,"path","total_conversions") heuristic_models(Data,"path","total_conversions",var_value="total_conversion_value") ## End(Not run)
Estimate a k-order Markov model from customer journey data. Differently from markov_model, this function iterates estimation until convergence is reached and enables multiprocessing.
markov_model(Data, var_path, var_conv, var_value=NULL, var_null=NULL, order=1, nsim_start=1e5, max_step=NULL, out_more=FALSE, sep=">", ncore=1, nfold=10, seed=0, conv_par=0.05, rate_step_sim=1.5, verbose=TRUE, flg_adv=TRUE)markov_model(Data, var_path, var_conv, var_value=NULL, var_null=NULL, order=1, nsim_start=1e5, max_step=NULL, out_more=FALSE, sep=">", ncore=1, nfold=10, seed=0, conv_par=0.05, rate_step_sim=1.5, verbose=TRUE, flg_adv=TRUE)
Data |
data.frame containing customer journeys data. |
var_path |
column name containing paths. |
var_conv |
column name containing total conversions. |
var_value |
column name containing total conversion value. |
var_null |
column name containing total paths that do not lead to conversions. |
order |
Markov Model order. |
nsim_start |
minimum number of simulations used in computation. |
max_step |
maximum number of steps for a single simulated path. if NULL, it is the maximum number of steps found into Data. |
out_more |
if TRUE, transition probabilities between channels and removal effects will be returned. |
sep |
separator between the channels. |
ncore |
number of threads used in computation. |
nfold |
how many repetitions are used to verify if convergence has been reached at each iteration. |
seed |
random seed. Giving this parameter the same value over different runs guarantees that results will not vary. |
conv_par |
convergence parameter for the algorithm. The estimation process ends when the percentage of variation of the results over different repetions is less than convergence parameter. |
rate_step_sim |
number of simulations used at each iteration is equal to the number of simulations used at previous iteration multiplied by rate_step_sim. |
verbose |
if TRUE, additional information about process convergence will be shown. |
flg_adv |
if TRUE, ChannelAttribution Pro banner is printed. |
An object of class data.frame with the estimated number of conversions and the estimated conversion value attributed to each channel.
Davide Altomare ([email protected]).
## Not run: library(ChannelAttribution) data(PathData) #Estimate a Makov model using total conversions markov_model(Data, var_path="path", "total_conversions") #Estimate a Makov model using total conversions and revenues markov_model(Data, "path", "total_conversions", var_value="total_conversion_value") #Estimate a Makov model using total conversions, revenues and paths that do not lead to conversions markov_model(Data, "path", "total_conversions", var_value="total_conversion_value", var_null="total_null") #Estimate a Makov model returning transition matrix and removal effects markov_model(Data, "path", "total_conversions", var_value="total_conversion_value", var_null="total_null", out_more=TRUE) #Estimate a Markov model using 4 threads markov_model(Data, "path", "total_conversions", var_value="total_conversion_value", ncore=4) ## End(Not run)## Not run: library(ChannelAttribution) data(PathData) #Estimate a Makov model using total conversions markov_model(Data, var_path="path", "total_conversions") #Estimate a Makov model using total conversions and revenues markov_model(Data, "path", "total_conversions", var_value="total_conversion_value") #Estimate a Makov model using total conversions, revenues and paths that do not lead to conversions markov_model(Data, "path", "total_conversions", var_value="total_conversion_value", var_null="total_null") #Estimate a Makov model returning transition matrix and removal effects markov_model(Data, "path", "total_conversions", var_value="total_conversion_value", var_null="total_null", out_more=TRUE) #Estimate a Markov model using 4 threads markov_model(Data, "path", "total_conversions", var_value="total_conversion_value", ncore=4) ## End(Not run)
Estimate a k-order transition matrix from customer journey data.
transition_matrix(Data, var_path, var_conv, var_null, order=1, sep=">", flg_equal=TRUE, flg_adv=TRUE)transition_matrix(Data, var_path, var_conv, var_null, order=1, sep=">", flg_equal=TRUE, flg_adv=TRUE)
Data |
data.frame containing customer journeys data. |
var_path |
column name containing paths. |
var_conv |
column name containing total conversions. |
var_null |
column name containing paths that do not lead to conversions. |
order |
Markov Model order. |
sep |
separator between the channels. |
flg_equal |
if TRUE, transitions from a channel to itself will be considered. |
flg_adv |
if TRUE, ChannelAttribution Pro banner is printed. |
An object of class List containing a dataframe with channel names and a dataframe with the estimated transition matrix.
Davide Altomare ([email protected]).
## Not run: library(ChannelAttribution) data(PathData) transition_matrix(Data, var_path="path", var_conv="total_conversions", var_null="total_null", order=1, sep=">", flg_equal=TRUE) transition_matrix(Data, var_path="path", var_conv="total_conversions", var_null="total_null", order=3, sep=">", flg_equal=TRUE) ## End(Not run)## Not run: library(ChannelAttribution) data(PathData) transition_matrix(Data, var_path="path", var_conv="total_conversions", var_null="total_null", order=1, sep=">", flg_equal=TRUE) transition_matrix(Data, var_path="path", var_conv="total_conversions", var_null="total_null", order=3, sep=">", flg_equal=TRUE) ## End(Not run)