System and apparatus for models based on data augmented with conceivable transitions

1. A system comprising:

an input/output interface to communicate with a client device;

a processor readable non-transitory data storage medium; and

a processor, in communication with the processor readable data non-transitory storage medium and the input/output interface, to:

augment historical interaction data of prior buyers and non-buyers that browsed a website by imputing connections among links between URLs of the website and estimating an augmented set of transition probabilities based on conceivable transitions arising from the imputed connections, each of the conceivable transitions including a multi-step transition between a first URL and a second URL via at least one intermediate URL of the website;

train each of a plurality of different models using the historical interaction data and/or augmented historical interaction data;

receive from a client device, interaction data of a user browsing the website;

provide a plurality of different models for an intent of the user based on the interaction data, the plurality of different models comprising at least one augmented buyer model based on the augmented set of transition probabilities of the conceivable transitions and an alternate buyer model based on an actual transition between the first URL and the second URL;

compare a transition probability of the augmented set of transition probabilities associated with the augmented buyer model based on a conceivable transition to a significance threshold measure;

when the transition probability of the augmented set of transition probabilities exceeds the significance threshold measure, select the augmented buyer model;

when the transition probability of the augmented set of transition probabilities does not exceed the significance threshold measure, select the alternate buyer model;

determine when the user is likely to be a buyer, based on the interaction data and the selected one of the augmented buyer model and the alternate buyer model; and

present the user with an offer to buy from the website upon the determination that the user is likely to be the buyer.

2. The system of claim 1 , wherein the augmented set of transition probabilities are modeled as a Markov chain of higher than a first order and wherein the processor is to:

determine whether the interaction data matches a buyer or nonbuyer model;

when the interaction data matches the nonbuyer model, repeat iteratively and as the user browses the website, the receiving of interaction data and determining whether the interaction data matches the buyer or the nonbuyer model; and

when the interaction data matches the buyer model, perform the comparing of the transition probability associated with the augmented buyer model and determining when the user is likely to be the buyer or a non-buyer.

3. The system of claim 1 , wherein the alternate buyer model is without augmentation, wherein each conceivable transition is a transition that could have occurred based on transitions from the first URL to the at least one intermediate URL and from the at least one intermediate URL to the second URL recorded in the historical interaction data, wherein, in the augmented historical interaction data, the processor models a sequence of clicks in the historical interaction data as a connected graph in which each of a plurality of nodes visited by a prior user to the website represents a URL and an edge between a selected pair of nodes are associated with the transition probability of a transition between the selected pair of nodes, and wherein in the comparing of the transition probability associated with the augmented buyer model based on a conceivable transition to the significance threshold measure comprises the processor determining whether a transition probability for any edge in the connected graph exceeds the significance threshold measure, selecting the augmented buyer model when the transition probability for one or more edges in the connected graph exceeds the significance threshold measure, and selecting the alternate buyer model when the transition probability for any edge in the connected graph does not exceed the significance threshold measure.

4. The system of claim 1 , wherein the alternate buyer model does not include any conceivable transition, wherein the processor augments the historical interaction data by building a second order or higher Markov chain, and wherein to determine if the user is likely to be the buyer the processor is to:

record transitions of the user between the URLs of the website as the interaction data; and

construct a transition matrix for the user based on the recorded transitions of the user.

5. The system of claim 4 , wherein the processor is to:

compare a transition probability in the transition matrix of the user with transition probabilities of the augmented set of transition probabilities from prior buyers and prior non-buyers.

6. The system of claim 1 , wherein the determination regarding the user likely being the buyer is based on the comparison of the transition probability associated with the augmented buyer model based on the conceivable transition to the significance threshold measure.

7. The system of claim 1 , further comprising:

receiving the historical interaction data of users of the website, the historical interaction data comprising one or more historical actual transitions between the first URL and the second URL;

build the alternate buyer model corresponding to the one or more historical actual transitions, the alternate buyer model having a first set of transition probabilities different from the augmented set of transition probabilities;

determine, based on the one or more historical actual transitions, one or more conceivable transitions between the first URL and the second URL; and

build the augmented buyer model based on the historical actual transition and the one or more conceivable transitions such that edges connecting nodes corresponding to the first URL and the second URL include the transition probabilities of the first set of transition probabilities and the augmented set of transition probabilities.

8. The system of claim 7 , wherein to train each of [a] the plurality of different models using the historical interaction data and/or the augmented historical interaction data the processor is to:

train the augmented buyer model using the augmented historical interaction data; and

train the alternate buyer model using the historical interaction data; and

wherein to select one of the plurality of different models the processor is to:

select one of the alternate buyer model, the augmented buyer model, and a non-buyer model.

9. The system of claim 8 , wherein, in the comparing of the transition probability associated with the augmented buyer model based on the conceivable transition to the significance threshold measure, the augmented set of transition probabilities comprises transition probabilities associated with each edge connecting nodes in a graph including the conceivable transitions and wherein the processor compares a transition probability of each edge with the significance threshold measure.

10. The system of claim 9 , wherein the processor is to:

determine minimum and maximum transition probabilities from a selected graph without augmentation; and

derive the significance threshold measure from the minimum and the maximum transition probabilities.

11. A computing apparatus comprising:

an input/output interface to communicate with a client device;

a processor readable data non-transitory storage medium; and

a processor, in communication with the processor readable data non-transitory storage medium and the input/output interface, to:

access historical interaction data;

identify buyer sessions from the historical interaction data;

augment the historical interaction data of the identified buyer sessions that browsed a website by imputing connections among links between URLs and estimating a set of transition probabilities based on conceivable transitions arising from the imputed connections that could have occurred in the historical interaction data, each of the conceivable transitions including a multi-step transition between a first URL and a second URL via at least one intermediate URL of the website;

build a buyer model and an augmented buyer model from the buyer sessions of the historical interaction data and/or augmented historical interaction data, the augmented buyer model based on the conceivable transitions and the buyer model based on actual transitions that occurred in the historical interaction data;

build a non-buyer model from non-buyer browsing sessions of the historical interaction data; and

provide the buyer model, the augmented buyer model, and the non-buyer model to a webserver for identifying buyers.

12. The computing apparatus of claim 11 , wherein the set of transition probabilities are modeled as a Markov chain of higher than a first order, wherein the buyer model is without augmentation, and wherein the processor:

trains the augmented buyer model using the augmented historical interaction data; and

trains the buyer model using the historical interaction data.

13. The computing apparatus of claim 11 , wherein the non-buyer model does not include any conceivable transition, wherein the processor augments the historical interaction data by building a second order or higher Markov chain, and wherein the processor further:

receives further historical interaction data of visitors to a website associated with the webserver;

provides the further historical interaction data for training the buyer model, the augmented buyer model, and the non-buyer model.

14. The computing apparatus of claim 13 , wherein the processor selects one of the buyer model, the augmented buyer model, and the non-buyer model according to the following rules:

the processor selects the non-buyer model for making a prediction regarding a user when interaction data is indicative of the user browsing without executing a purchasing transaction; and

the processor selects one of the buyer model or the augmented buyer model based on a comparison of transition probabilities of the augmented buyer model with a threshold measure for making a prediction regarding the user if the interaction data is indicative of the user making the purchasing transaction, wherein, in the augmented historical interaction data, the processor models a sequence of clicks in the historical interaction data as a connected graph in which each of a plurality of nodes visited by a prior user to the website represents a URL and an edge between a selected pair of nodes is associated with the transition probability of a transition between the selected pair of nodes, and wherein in the comparing of the transition probability associated with the augmented buyer model based on a conceivable transition to the threshold measure comprises the processor determining whether a transition probability for any edge in the connected graph exceeds the threshold measure, selecting the augmented buyer model when the transition probability for one or more edges in the connected graph exceeds the threshold measure, and selecting the buyer model when the transition probability for any edge in the connected graph does not exceed the threshold measure.

15. The computing apparatus of claim 11 , wherein the processor repeats iteratively upon a determination that a user is likely a non-buyer and as the user browses the website, a receipt, from the client device, of interaction data of a user browsing the website and a determination when the user is likely to be a buyer or the non-buyer, based on the interaction data of the user browsing the website and a selected model determining if the user is likely to be the buyer or the non-buyer.

16. The computing apparatus of claim 11 , wherein each conceivable transition is a transition that could have occurred based on transitions from the first URL to at least one intermediate node and from the at least one intermediate node to the second URL recorded in the historical interaction data.

17. A web server comprising:

an input/output interface to communicate with a client device;

a processor readable non-transitory data storage medium; and

a processor, in communication with the processor readable non-transitory data storage medium and the input/output interface, wherein the processor readable non-transitory data storage medium comprises:

sessions data of different prior buyers and non-buyers who browsed a website;

augmented historical interaction data of prior buyers and non-buyers that browsed the website by imputing connections among links between URLs, the augmented historical interaction data comprising augmented sessions data modeling multi-step conceivable transitions that could have occurred in the sessions data as a connected graph in which each of a plurality of nodes visited by a selected prior buyer or non-buyer to the website represents a URL and an edge between a selected pair of nodes is associated with a transition probability in an augmented set of transition probabilities of a transition between the selected pair of nodes;

a plurality of different models, executable by the processor, comprising a buyer model without augmentation, an augmented buyer model, and a non-buyer model for determining a buying intent of a website visitor, wherein:

the plurality of different models are trained on the sessions data and augmented sessions data, and

the buyer model without augmentation is based on transition probabilities of actual transitions that occurred in historical interaction data, the augmented buyer model is based on transition probabilities, and the non-buyer model is based on transition probabilities of browsing sessions that do not include a purchasing transaction, and

wherein the processor:

selects one of the plurality of different models based on a comparison of each of the plurality of different models with interaction data received from the client device of a user visiting the website, wherein the comparison comprises the processor determining whether a transition probability for any edge in the connected graph exceeds a significance threshold measure, selecting the augmented buyer model when the transition probability for one or more edges in the connected graph exceeds the significance threshold measure, and selecting the buyer model or non-buyer model when the transition probability for any edge of the one or more edges in the connected graph does not exceed the significance threshold measure; and

executes an action based on the selected one of the plurality of different models.

18. The web server of claim 17 , wherein the processor selects one of the plurality of different models, selects the non-buyer model for making a prediction regarding the user when the interaction data is indicative of the user browsing without executing the purchasing transaction; and selects one of the buyer model without augmentation or the augmented buyer model based on comparison of transition probabilities for one or more edges in the connected graph if the interaction data is indicative of the user making the purchasing transaction.

19. The web server of claim 17 , wherein the buyer model and the non-buyer model do not include any conceivable transition, wherein the processor augments the sessions data by building a second order or higher Markov chain, and wherein in building the buyer model without augmentation the processor builds the buyer model from actual transitions in the historical interaction data.

20. The web server of claim 17 , wherein the processor further:

receives interaction data of visitors to the website associated with the web server; and

provides the interaction data for training the buyer model without augmentation, the augmented buyer model, and the non-buyer model.