System, method and computer program product for predicting item preference using revenue-weighted collaborative filter
Embodiments disclosed provide a system, method, and computer program product for identifying consumer items more likely to be bought by an individual user. In some embodiments, a collaborative filter may be used to rank items based on the degree to which they match user preferences. The collaborative filter may be hierarchical and may take various factors into consideration. Example factors may include the similarity among items based on observable features, a summary of aggregate online search behavior across multiple users, the item features determined to be most important to the individual user, and a baseline item against which a conditional probability of another item being selected is measured.
1. A method for identifying consumer items more likely to be bought by an individual user, comprising:
determining, by a vehicle data system having a processor and a memory, for each individual item of a plurality of items, a probability that the individual user will select the individual item, given that the individual user has expressed interest in a baseline item, wherein the probability is determined based on:
a similarity among the plurality of items based on observable features thereof;
an aggregate online search behavior across multiple users; and
preferences of the individual user in the baseline item, wherein the baseline item is established by the individual user; and
ranking, in decreasing order, probabilities determined for the plurality of items, wherein items that are more likely to be bought by the individual user are ranked higher than items that are less likely to be bought by the individual user;
wherein the method further comprises:
for each paired observation associated with a baseline item i established by the individual user, determining what a first next item j is expected to be selected;
examining all observations associated with the baseline item i to determine what a second next item t is expected to be selected;
making a prediction as to which next item the individual user will select;
comparing the prediction with an actual next item k selected by the individual user; and
assigning a penalty value to the prediction if the prediction is incorrect.
2. A method according to claim 1 , further comprising determining the similarity among the plurality of items:
a) computing individual feature difference between a first observation and a second observation;
b) computing a composite similarity between the first observation and the second observation; and
c) repeating a) and b) for all possible values of the first observation and the second observation.
3. A method according to claim 1 , further comprising collecting item view frequencies only for each hop by the individual user in a sequence of item discovery.
4. A method according to claim 1 , further comprising collecting item view frequencies only for all hops by the individual user in a sequence of item discovery.
5. A method according to claim 1 , further comprising collecting item view frequencies only for all pairs of items in a sequence of item discovery.
6. A method according to claim 1 , further comprising:
defining a kernel for a given radius each time the second next item t is selected after the baseline item i.
7. A method according to claim 1 , further comprising:
determining a conditional probability that the individual user will select the first next item j after the baseline item i.
8. A method according to claim 1 , wherein the prediction is one of a plurality of predictions, further comprising:
determining a total penalty value for all incorrect predictions in the plurality of predictions; and
determining a set of weights that minimizes the total penalty.
9. A computer program product comprising at least one non-transitory computer readable medium storing instructions translatable by at least one processor to implement a collaborative filter, said collaborative filter comprising:
a first component configured to determine a similarity among a plurality of items based on observable features thereof;
a second component configured to aggregate online search behavior across multiple users; and
a third component configured to determine what features of the observable features are most important to an individual user, wherein the third component of the collaborative filter examines item discovery behavior of the individual user across multiple dimensions with respect to the observable features in a baseline item and at least one next item, wherein the baseline item is established by the individual user,
wherein a probability that the individual user will select a next item, given that the user has expressed interest in the base line item, is determined based at least in part on the similarity among the plurality of items and the item discovery behavior of the individual user relative to the aggregate online search behavior across multiple users, and
wherein the instructions are further translatable by the at least one processor to:
for each paired observation associated with a baseline item i established by the individual user, determine what a first next item j is expected to be selected;
examine all observations associated with the baseline item i to determine what a second next item t is expected to be selected;
make a prediction as to which next item the individual user will select;
compare the prediction with an actual next item k selected by the individual user; and
assign a penalty value to the prediction if the prediction is incorrect.
10. A computer program product according to claim 9 , wherein the first component of the collaborative filter is further configured to:
a) compute individual feature difference between a first observation and a second observation;
b) compute a composite similarity between the first observation and the second observation; and
c) repeat a) and b) for all possible values of the first observation and the second observation.
11. A computer program product according to claim 9 , wherein the second component of the collaborative filter is further configured to collect item view frequencies only for each hop by the individual user in a sequence of item discovery.
12. A computer program product according to claim 9 , wherein the second component of the collaborative filter is further configured to collect item view frequencies only for all hops by the individual user in a sequence of item discovery.
13. A computer program product according to claim 9 , wherein the second component of the collaborative filter is further configured to collect item view frequencies only for all pairs of items in a sequence of item discovery.
14. A system, comprising:
a processor; and
a memory storing instructions translatable by the processor to implement a collaborative filter having:
a first software component configured to determine a similarity among a plurality of items based on observable features thereof;
a second software component configured to aggregate online search behavior across multiple users; and
a third software component configured to determine what features of the observable features are most important to an individual user, wherein the third component of the collaborative filter examines item discovery behavior of the individual user across multiple dimensions with respect to the observable features in a baseline item and at least one next item, wherein the baseline item is established by the individual user,
wherein the instructions are further translatable by the processor to:
for each paired observation associated with a baseline item i established by the individual user, determine what a first next item j is expected to be selected;
examine all observations associated with the baseline item i to determine what a second next item t is expected to be selected;
make a prediction as to which next item the individual user will select;
compare the prediction with an actual next item k selected by the individual user; and
assign a penalty value to the prediction if the prediction is incorrect.
15. A system according to claim 14 , wherein the first software component of the collaborative filter is further configured to:
a) compute individual feature difference between a first observation and a second observation;
b) compute a composite similarity between the first observation and the second observation; and
c) repeat a) and b) for all possible values of the first observation and the second observation.
16. A system according to claim 14 , wherein the second software component of the collaborative filter is further configured to collect item view frequencies only for each hop by the individual user in a sequence of item discovery.
17. A system according to claim 14 , wherein the second software component of the collaborative filter is further configured to collect item view frequencies only for all hops by the individual user in a sequence of item discovery.
18. A system according to claim 14 , wherein the second software component of the collaborative filter is further configured to collect item view frequencies only for all pairs of items in a sequence of item discovery.