Predictive modeling system applied to contextual commerce
An automated method, non-transitory computer-readable storage device and system for developing predictive models including predictive causal models and using said models to develop a personalized context for use in advertising, configuring, offering, producing, and/or delivering offerings that are appropriate to the context of a specific individual, group or organization.
1 . A non-transitory computer-readable storage device encoded with a computer program product, the computer program product comprising instructions that when executed by one or more computers cause the one or more computers to perform operation comprising:
aggregating a plurality of data related to a user entity and to one or more offerings that may be provided by one or more offering entities to said user entity in a format suitable for automated analysis;
transforming at least a portion of said data into a context for said user entity where said context comprises a plurality of layers wherein at least one layer comprises a predictive model developed by learning from at least a portion of the data;
using said context and the data for the one or more offerings to create a personalized offering for one or more steps in a commerce chain
where the one or more steps in a commerce chain are selected from the group consisting of advertise, configure, produce, offer and deliver.
2 . The non-transitory computer-readable storage device of claim 1 , wherein the personalized offering comprises an optimal offering for the user entity, the offering entity or for a combination thereof.
3 . The non-transitory computer-readable storage device of claim 1 , wherein the personalized offering is selected from the group consisting of ad, configuration, data, information, knowledge, media, product, service, offer term, offer condition, delivery mode, delivery time and delivery location.
4 . The non-transitory computer-readable storage device of claim 1 , wherein the personalized offering is delivered as a time when the user entity is most likely to be receptive to an interruption.
5 . The non-transitory computer-readable storage device of claim 1 , wherein the personalized offering is delivered as required to support an upcoming decision.
6 . The non-transitory computer-readable storage device of claim 1 , wherein the personalized offering is delivered when the user context matches a pre-defined context or when a keyword is entered into a search.
7 . The non-transitory computer-readable storage device of claim 1 , wherein developing the predictive model for the at least one context layer by learning from at least the portion of said data comprises:
using a plurality of different types of predictive models and a plurality of different types of causal models to analyze and select the portion of the data to use as an input when modeling the at least one context layer;
learning which predictive model type from the plurality of different types of predictive models to include in the predictive model for the at least one context layer when using the selected data;
learning which causal model type from the plurality of the different types of causal models comprises a best fit for modeling the at least one context layer when using the selected data; and
learning if a clustering of the selected input data improves an accuracy of the predictive model for the at least one context layer where the plurality of different types of causal models are selected from the group consisting of Tetrad, LaGrange, Bayesian, probabilistic relational and path analysis and where the plurality of different types of predictive models are selected from the group consisting of classification and regression tree, generalized autoregressive conditional heteroskedasticity, projection pursuit regression, stepwise regression, logistic regression, probit regression, factor analysis, growth modeling, linear regression; redundant regression network, boosted Naive Bayes Regression, support vector method, markov models, kriging, multivalent models, relevance vector method, multivariate adaptive regression splines, rough-set analysis, generalized additive model and stepwise regression.
8 . A system comprising: one or more computers; and one or more data storage devices having instructions stored thereon that, when executed by the computers, cause the computers to perform operations comprising:
training each of a plurality of different types of predictive models using training data to analyze and select a portion of the training data to use as an input to a next stage of modeling;
learning if a clustering of the selected portion of the training data improves an accuracy of any of the different types of predictive models;
learning which model from a plurality of causal models comprises a best fit model when using the selected portion of the training data and then refining the selected portion of the training data to include only the data selected by the best fit causal model where said refined selection of the training data comprises the refined training data; and
outputting the best fit causal model where the best fit causal model comprises a predictive causal model, where the plurality of causal models are selected from the group consisting of Tetrad, LaGrange, Bayesian, probabilistic relational and path analysis and where the plurality of different types of predictive models are selected from the group consisting of classification and regression tree, generalized autoregressive conditional heteroskedasticity, projection pursuit regression; stepwise regression, logistic regression, probit regression, factor analysis, growth modeling, linear regression, redundant regression network, boosted Naive Bayes Regression, support vector method, markov models, kriging, multivalent models, relevance vector method, multivariate adaptive regression splines, rough-set analysis, generalized additive model and stepwise regression.
9 . The system of claim 8 , wherein the operations further comprise:
training each of the plurality of the different types of predictive models using training data, wherein the predictive models include a plurality of each type of predictive model that are trained with different combinations of features of the training data;
generating, for each of the plurality of trained predictive models, a measure that represents an estimation of an effectiveness of the respective trained predictive models; and
selecting one predictive model from the plurality of different types of predictive models for output as the final predictive model based on the respective measures of the trained predictive models.
10 . The system of claim 9 , wherein the measure that represents the estimation of the effectiveness of the respective trained predictive models comprises a mean squared error measure.
11 . The system of claim 8 , wherein learning which model from the plurality of causal models comprises the best fit model when using the selected portion of the training data comprises using a cross validation algorithm to identify the best fit model.
12 . The system of claim 8 , wherein learning if the clustering of the selected portion of the training data improves the accuracy of any of the predictive models comprises comparing an error measure for an overall model with a combined error measure from models of two or more clusters.
13 . The system of claim 8 , wherein the training data are clustered using one or more algorithms selected from the group consisting of unsupervised “Kohonen” neural network, decision tree, support vector method, K-nearest neighbor, expectation maximization (EM) and the segmental K-means algorithm.
14 . A system comprising:
one or more computers; and one or more data storage devices having instructions stored thereon that, when executed by the computers, cause the computers to perform operations comprising:
aggregate a plurality of data related to a user entity and to one or more offerings that may be provided by one or more offering entities to said user entity in a format suitable for automated analysis;
transform at least a portion of said data into a context for said user entity where said context comprises a plurality of layers wherein at least on layer comprises a predictive model developed by learning from at least a portion of the data;
use said context and the data for the one or more offerings to create a personalized offering for one or more steps in a commerce chain
where the one or more steps in a commerce chain are selected from the group consisting of advertise, configure, produce, offer and deliver.
15 . The system of claim 14 , wherein the personalized offering comprises an optimal offering for the user entity, the offering entity or for a combination thereof.
16 . The system of claim 14 , wherein the personalized offering is selected from the group consisting of ad, configuration, data, information, knowledge, media, product, service, offer term, offer condition, delivery mode, delivery time and delivery location.
17 . The system of claim 14 , wherein the personalized offering is delivered as a time when the user entity is most likely to be receptive to an interruption.
18 . The system of claim 14 , wherein the personalized offering is delivered as required to support an upcoming decision or when a keyword is entered into a search.
19 . The system of claim 14 , wherein the personalized offering is delivered when the user context matches a pre-defined context.
20 . The system of claim 14 , wherein developing the predictive model for the at least one context layer by learning from at least the portion of said data comprises:
using a plurality of different types of predictive models and a plurality of different types of causal models to analyze and select the portion of the data to use as an input when modeling the at least one context layer;
learning which predictive model type from the plurality of different types of predictive models to include in the predictive model for the at least one context layer when using the selected data;
learning which causal model type from the plurality of the different types of causal models comprises a best fit for modeling the at least one context layer when using the selected data; and
learning if a clustering of the selected input data improves an accuracy of the predictive model for the at least one context layer where the plurality of different types of causal models are selected from the group consisting of Tetrad, LaGrange, Bayesian, probabilistic relational and path analysis and where the plurality of different types of predictive models are selected from the group consisting of classification and regression tree, generalized autoregressive conditional heteroskedasticity, projection pursuit regression, stepwise regression, logistic regression, probit regression, factor analysis, growth modeling, linear regression; redundant regression network, boosted Naive Bayes Regression, support vector method, markov models, kriging, multivalent models, relevance vector method, multivariate adaptive regression splines, rough-set analysis, generalized additive model and stepwise regression.