Computer graphics processing and selective visual display system – Display driving control circuitry – Controlling the condition of display elements
Reexamination Certificate
2001-05-24
2004-12-14
Kincaid, Kristine (Department: 2174)
Computer graphics processing and selective visual display system
Display driving control circuitry
Controlling the condition of display elements
C705S001100, C705S005000, C705S014270, C705S014270, C705S026640, C705S027200, C707S793000, C707S793000, C707S793000, C707S793000, C707S793000
Reexamination Certificate
active
06831663
ABSTRACT:
BACKGROUND OF INVENTION
1. Technical Field
The present invention involves a new system and process for automatically assigning scores to predictor values for measuring the influence of each predictor/variable value pair on a prediction of likely user choices.
2. Related Art
By way of background, collaborative filtering or recommender systems typically use various probabilistic methods in combination with one or more databases comprised of user preferences to predict additional topics, items, or products a new user might like. Generally, the goal of such systems is to predict the utility of items to a particular user based on a database of user preferences or votes from a sample or population of other users. Either of two general classes of collaborative filtering algorithms, e.g., memory-based algorithms or model-based collaborative filtering, is often used for making such predictions. One limitation of such systems is that while they are useful for predicting user preferences, they are not useful for determining the contribution or influence of particular user preferences or votes from the population of users on particular user preference predictions.
For example, a probabilistic model can be used for making predictions in a movie-recommendation application that suggests movies to users based on the other movies that they have seen. One method of implementing such a system is to build a probabilistic model over the set of variables M={M
1
, . . . , M
n
}, where M
i
is a binary variable denoting whether or not a user has seen the ith movie. The model is constructed such that, for each movie i, the conditional probability, p(M
i
=watched |M\{M
i
}) is extracted for each movie. In general, such a system is used to recommend movies as follows: for each movie M
i
that a user has not watched, the probability that the user would have watched the movie is calculated based on the assumption that it is unknown whether the user has actually watched movie M
i
. Such a system then recommends those movies that have the highest posterior probability. However, while this system recommends movies based on the probability that a user will want to watch the movies, it fails to answer the question as to why the user might want to watch the movies. In other words, such a system is unable to determine which other movies from the set M={M
1
, . . . , M
n
} were most influential in making the particular probabilistic recommendations to the user.
Consequently, what is needed is a system and process for determining which elements of a probabilistic model have the greatest influence on particular probabilistic predictions or recommendations computed from the probabilistic model. Determining which elements have the greatest influence on the predictions or recommendations allows the question to be answered as to why such predictions or recommendations were made.
SUMMARY OF INVENTION
In general, the present invention solves the aforementioned problems, as well as other problems that will become apparent from an understanding of the following description by automatically assigning “scores” to the predictor/variable value pairs of a conventional probabilistic model to measure the relative impact or influence of particular elements of a set of topics, items, products, etc. on particular predictions. In particular, these scores measure the relative impact, either positive or negative, that the value of each individual predictor variable has on the posterior distribution of the target topic, item, product, etc., for which a probability is being determined. These scores are useful for understanding why each prediction is made, and how much impact each predictor has on the prediction. Consequently, such scores are useful for explaining why a particular recommendation was made.
For example, where a probabilistic model recommends particular movies to a user based upon other movies that the user has seen, an application embodying the present invention provides information as to what other movies were most influential in making the particular recommendations. Thus, for example, where the predictor/variable value pairs associated with watching movie j, and with not watching movie k, have the highest scores, these predictor/variable value pairs are provided as the most influential factors in determining whether the user should be provided with a recommendation to watch movie i. In one embodiment, such information is provided as a human-readable or “natural language” explanation such as, for example, “Movie i was recommended to the user because the user watched movie j, but did not watch movie k.” Further, any number of the top most influential topics, items, products, etc., may be provided for the purpose of explaining particular recommendations. In other words, the top n most influential predictor/variable value pairs can be provided for the purpose of explaining particular probabilistic recommendations. Clearly, the most positive influences as well as the most negative influences may be identified.
In accordance with preceding discussion, a system and method according to the present invention operates to automatically assign scores to members of a set of at least one predictor/variable value pair representing likely user choices for determining the effect or influence of those predictor/variable value pairs on predictions of one or more likely user choices. This is accomplished by first obtaining a set of user preferences for a particular user. These preferences are obtained either explicitly or implicitly using conventional techniques.
For example, one common method for explicitly obtaining user preferences involves asking a user to rate particular objects, such as topics, items, products, books, movies, food, drinks, etc., on some predefined schedule or list. One example of implicitly obtaining user preferences is to simply identify whether a user has used, purchased, viewed, etc., such objects. Further, another common method for implicitly obtaining user preferences involves observing user behavior in order to impute particular preferences to that user. For example, the idea here is that by watching a particular movie, the user is showing a preference for that movie over the movies that the user didn't watch. Examples of such observation include observing a user's Internet web browsing behavior, i.e., what items or objects does the user look at while browsing; observing a user's purchasing history to see what the user buys, when the user buys, and where the user buys; and observing other available informational patterns. Clearly, any conventional technique for either implicitly or explicitly determining user preferences, or some combination of both implicit and explicit determination of user preferences, may be used to determine a set of user preferences.
Next, once the user preferences have been determined using conventional techniques, the probability of each variable/variable value pair is computed, again using conventional techniques using the set of user preferences in combination with a probabilistic model. Simple examples of variable/variable value pairs include “Movie 1/watched,” or “Book 5
ot read.” The probabilistic model can be any conventional type, including, for example, dependency networks and Bayesian networks, so long as the probabilistic model can be used to compute the probability of variable/variable value pairs. Next, at least one prediction of likely user choices is automatically computed based upon the variable/variable value pairs using conventional techniques.
To differentiate the variable/variable value pairs from the predictions of likely user choices, the term “predictor/variable value pair” is used in place of “variable/variable value pair” to refer to a variable that is used in the probabilistic model to predict the specific target for which a prediction is being made. Further, it should be noted that the set of all predictor/variable value pairs that are evaluated is preferably restricted by those pair
Chickering David Maxwell
Heckerman David E.
Rounthwaite Robert
Kincaid Kristine
Lyon & Harr LLP
Microsoft Corporation
Pesin Boris
Watson Mark A.
LandOfFree
System and process for automatically explaining... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with System and process for automatically explaining..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and process for automatically explaining... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3307125