Data processing: measuring – calibrating – or testing – Measurement system – Measured signal processing
Reexamination Certificate
1999-06-29
2001-11-20
Hoff, Marc S. (Department: 2857)
Data processing: measuring, calibrating, or testing
Measurement system
Measured signal processing
C705S014270
Reexamination Certificate
active
06321179
ABSTRACT:
FIELD OF THE INVENTION
This invention relates generally to collaborative filtering systems for predicting a user's level of interest in information, and more particularly to a collaborative filtering system that provides users with an estimate of the accuracy of its predictions.
BACKGROUND OF THE INVENTION
The amount of information that is available globally, via the World Wide Web or the Internet, or locally on some Intranets, is so large that managing such information is critical. One way of managing and distributing information is to use a collaborative filtering system to predict a user's preference and use that information to distribute new information to the user.
Collaborative filtering, the sharing of knowledge through recommendations, is an important vehicle for distributing information. There are two distinct types of collaborative filtering mechanisms: those which enable active collaborative filtering by making it easier for people to share pointers to interesting documents and those which automate collaborative filtering by using statistical algorithms to make recommendations based on correlations between personal preferences. Collaborative filtering is used in many recommender systems, i.e., systems which provide recommendations to users.
Automated collaborative filtering (ACF) is a general type of statistical algorithm that matches items (such as movies, books, music, news articles, etc.) to users by first matching users to each other. ACF uses statistical algorithms to make recommendations based on correlations between personal preferences. Recommendations usually consist of numerical ratings input manually by users, but can also be deduced from user behavior (e.g., time spent reading a document, actions such as printing, saving or deleting a document). The premise of such systems is that a user is going to prefer an item that is similar to other items chosen by the user and by other users.
U.S. Pat. No. 5,724,567 to Rose et al. entitled “System for Directing Relevance Ranked Data Objects to Computer Users” describes a system for matching a user's interests by comparing the content of an item with a user's adaptive profile. Feedback is also available to enable the user to update his/her profile.
U.S. Pat. No. 5,704,017 to Heckerman et al. entitled “Collaborative Filtering Utilizing a Belief Network” describes a collaborative filtering system which employs a belief network to predict the preferences of a user using probabilistic inference. In performing probabilistic inference, the known attributes of a user are received and the belief network is accessed to determine the probability of the unknown preferences of the user given the known attributes. Based on these probabilities, the preference most likely to be desired by the user can be predicted.
Collaborative filtering systems are of particular value to suppliers of goods and services in that they can be used to enhance the distribution of their goods and services to customers. Current applications of ACF technology in electronic commerce include recommendation services given away freely by retailers to attract customers to their offerings and to provide more specialized services and knowledge management tools for organizations. For both kinds of applications, the quality of the recommendation service is important.
Automated collaborative filtering systems such as the above suffer from the cold-start problem: early users will receive inaccurate predictions until there is enough usage data for the algorithm to be able to learn their preferences. In applications of ACF technology, such as knowledge management tools for organizations, consistent high quality service is key. Many existing current systems which employ ACF (MovieLens, Arnazon.com, BarnesandNoble, etc.) either require users to rate a number of items before they will provide recommendations, use data from purchases, or provide initial predictions which are not personalized (e.g., use the average rating).
Even once a collaborative filtering system has learned something about a user, the statistical significance and accuracy of predictions made using ACF (whether using a correlation calculation or a probabilistic inference) for any given item varies widely. For example, it may vary depending on how many others have rated the item. It may also depend on the number of items rated in common by the users. From the user's perspective, feedback about the accuracy of the predictions can help them build up confidence in the system and better understand the significance of the predictions and how the prediction mechanism works.
Making predictions based on a large number of ratings or a large number of ratings by other users will generally increase the accuracy of the prediction, but can take significant processing time. Some users may want to have the option of specifying the accuracy criteria in order to save processing time. These users may not be willing to wait for more accurate predictions and are less likely to be tolerant of errors, knowing that they trade off accuracy for timeliness. Other users, who tend to act as information filters, or gatekeepers, may be more tolerant of errors in predictions. However, for a given domain of interest, there will be a group of users who prefer to wait for accurate, personalized predictions.
There is a need for a collaborative filtering system that provides users with an estimate of the accuracy of its predictions. There is a need for a system and method of providing predictions to users in which the users can select or filter the accuracy of the prediction in exchange for timeliness. There is a need for a collaborative filtering system and method that provides users with high quality recommendations. There is a need for a collaborative filtering system which can accommodate different levels of accuracy for different users.
SUMMARY OF THE INVENTION
A collaborative filtering system for predicting a user's rating for an item, according to the invention, includes a memory storing a set of correlation coefficients, where each correlation coefficient is a measure of similarity in ratings between pairs of users in the system who have rated a particular item, and ratings for the item made by other users in the system. A processor, responsive to a request for a predicted user's rating for an item, calculates the predicted user's rating. The predicted user's rating is the sum of the product of a rating and its respective correlation coefficient divided by the sum of the correlation coefficients. The processor also calculates the variance of the predicted user's rating, where the variance of the predicted user's rating is a function of the variance of the correlation coefficients and the variance of the ratings. The variance of the predicted user's rating gives the user a range of the accuracy of the prediction.
The collaborative filtering system according to the invention can be implemented in one of many ways. For example, the system can be implemented in software operated by a network server. Users may access the server through a remote work station and input their requests. The server may be part of an Intranet or Internet network. Users may also provide ratings to the system which are stored in memory and may be used to update the correlation coefficients, as well as the variance calculations. (In the inventors' copending application Ser. No. 09/342,862, the variance may be stored and also used for updating the correlation coefficients.) The system can also be implemented as a software program on a personal computer.
Specific relationships for calculating the predicted user's rating are provided, as well as specific relationships for calculating the variance of the predicted user's rating. Information concerning the accuracy (in the form of the variance) can then be provided to users to help them judge the predicted user's rating. This information can also be used to filter which items are presented to the user based on accuracy. Users can tune
Dardenne Manfred
Glance Natalie S.
Hoff Marc S.
Pretlow Demetrius
Walder Jeannette M.
Xerox Corporation
LandOfFree
System and method for using noisy collaborative filtering to... 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 method for using noisy collaborative filtering to..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method for using noisy collaborative filtering to... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2583155