Three-way media recommendation method and system
View Patent ↗An electronic programming guide (EPG) system employing a preference engine and processing system that combines explicit rule profile, history profile, and feedback profile data to generate new predictions. Television shows are presumed to be indexed by many features. These features are extracted and counted for TV shows watched (implicit profile), and for TV shows rated by the viewer (feedback profile). These profiles are straightforward to combine with suitably greater weight being given to the feedback information. In addition, explicit profiles can make recommendations that stand alone or may be used to modify recommendations arising from either of the two sources. The modifications may take the form of additive or multiplicative changes to the existing recommendations or some other suitable mathematical form.
1. An automated recommendation system, comprising
a processor connected to receive resource data defining available resources and at least two sets of profile data, each defining a user's preferences with respect to the resources;
each of the sets of profile data being derived from a different class of interaction of the user with a first portion of the resource data and usable to predict a given resource's desirability based on each of the sets;
the processor being adapted to:
generate at least two sets of predictions based on one or a combination of the sets of profile data, and
combine the predictions by weight-averaging corresponding ones from each of the at least two sets of predictions.
2. A system as in claim 1 , wherein
the processor is further adapted to:
generate a weighted sum of corresponding records from each of the sets of profile data to generate a single combined set of profile data, and
generate at least one of the sets of predictions from the single combined set.
3. A system as in claim 2 , wherein
the processor is connected to control a delivery of resources corresponding to the resource data and responsively to the predictions.
4. A system as in claim 1 , wherein
the processor is connected to control a delivery of resources corresponding to the resource data and responsively to the predictions.
5. A system as in claim 1 , wherein
the at least two profile data sets include
a feedback data set derived from ratings provided by the user with respect to a particular resource in the resource data.
6. A system as in claim 1 , wherein
the at least two profile data sets include
an implicit data set derived from machine-observation of a user's resource use history, whereby the implicit data reflects the user's selections of resources to use.
7. A system as in claim 1 , wherein
at least one set of the at least two profile data sets comprises input vectors, and
the input vectors each include feature-value pairs.
8. A system as in claim 1 , wherein
at least one set of the at least two profile data sets comprises input vectors, and
the input vectors include feature-value pairs and a rating value.
9. A method of recommending resources, comprising:
generating by a processing device, at least two sets of profile data based on expressed preferences of a user with respect to the resources, each being usable to predict a given resource's desirability based on each of the sets;
generating by a processing device, at least two sets of predictions based on one or a combination of the sets of profile data; and
combining, by a processing device, the predictions by weight-averaging corresponding ones from each of the at least two sets of predictions.
10. A method as in claim 9 , further comprising:
generating a weighted sum of corresponding records from each of the sets of profile data to generate a single combined set of profile data; and
generating at least one of the sets of predictions from the single combined set.
11. A method as in claim 10 , further comprising
controlling a delivery of resources corresponding to the resource data responsively to the predictions.
12. A method as in claim 9 , further comprising
controlling a delivery of resources corresponding to the resource data responsively to the predictions.
13. A method as in claim 9 , wherein
generating the at least two sets of profile data includes
generating a feedback data set by accepting ratings from the user with respect to a particular resource in the resource data.
14. A method as in claim 9 , wherein
generating the at least two sets of profile data includes
generating an implicit data set by observing the user's resource use history, whereby the implicit data reflects the user's selections of resources to use.
15. A method as in claim 9 , wherein
at least one set of the at least two sets of profile data comprises input vectors, and
the input vectors each include feature-value pairs.
16. A method as in claim 9 , wherein
at least one set of the at least two sets of profile data comprises input vectors, and
generating the at least two sets of profile data includes generating feature-value pairs and a rating value.
17. A method as in claim 9 , wherein:
the sets of profile data includes:
a set of explicit profile data indicating express indications by a user of preferred classes of programming rather than indications by the user of particular resources that are preferred;
a feedback data set derived from ratings provided by the user with respect to a particular resource in the resource data; and
an implicit data set derived from machine-observation of a user's resource use history, whereby the implicit data reflects the user's selection.
18. An automated recommendation system, comprising:
a processor connected to receive resource data defining available resources and sets of profile data, each defining a users preferences with respect to the resources;
the sets of profile data including:
a set of explicit profile data indicating express indications by a user of preferred classes of programming rather than indications by the user of particular resources that are preferred;
feedback data set derived from ratings provided by the user with respect to a particular resource in the resource data; and
an implicit data set derived from machine-observation of a users resource use history, whereby the implicit data reflects the user's selection;
the processor being adapted to generate at least two sets of predictions based on one or a combination of the sets of profile data, each of the predictions including a confidence level;
the processor being further adapted to combine the predictions by weight-averaging corresponding ones from each of the at least two sets of predictions.
19. A system as in claim 18 , wherein
the processor is further adapted to adjust weights of the weight averaging responsively to a difference between the corresponding ones.
20. A system as in claim 18 , wherein
the processor is further adapted to selectively override the weight averaging responsively to a difference between the corresponding ones.
21. A method of automatically recommending resources, comprising:
receiving resource data defining available resources and sets of profile data, each defining user preferences with respect to the resources;
the sets of profile data including:
a set of explicit profile data indicating express indications by a user of preferred classes of programming rather than indications by the user of particular resources that are preferred;
a feedback data set derived from ratings provided by the user with respect to a particular resource in the resource data; and
an implicit data set derived from machine-observation of a user's resource use history, whereby the implicit data reflects the user's selection;
generating at least two sets of predictions based on one or a combination of the sets of profile data, each of the predictions including a confidence level; and
combining the predictions by weight-averaging corresponding ones from each of the at least two sets of predictions to produce a combined set.
22. A method in claim 21 , wherein
combining the predictions includes
adjusting weights of the weight averaging responsively to a difference between the corresponding ones.
23. A method as in claim 21 , wherein
combining the predictions includes
selectively overriding the weight averaging responsively to a difference between the corresponding ones such that a prediction of a one of the sets of predictions is included in the combined set and a prediction of the other of the sets of predictions is excluded.
24. A method of combining profile data, comprising:
generating first profile data by receiving through a user interface user preferences in the form of expressed generalized preferences corresponding classes of resources;
generating second profile data by receiving user preferences in the form of rating data corresponding to specific resources; and
applying the first and second profile data to respective prediction engines to produce first and second prediction results and combining the first and second results.
25. A method as in claim 24 , further including
combining the first and second profile data,
wherein
combining the first and second profiles includes weight averaging corresponding ones of the profile data.
26. A method as in claim 24 , wherein
combining respective results includes selectively weight averaging corresponding ones of the predictions.