Method and apparatus for recommending selections based on preferences in a multi-user system

• US 5,583,763 A
• Filed: 09/09/1993
• Issued: 12/10/1996
• Est. Priority Date: 09/09/1993
• Status: Expired due to Term

• Current Assignee: Intel Corporation
• Original Assignee: MNI Interactive
• Inventors: Miller, James R. III, Atcheson, John
• Primary Examiner(s): Weinhardt, Robert A.
• Application Number:
  US08/119,793
• Time in Patent Office: 1,188 Days
• Field of Search: 364/401, 364/402, 364/419.2, 395/600
• US Class Current: 707/750
• CPC Class Codes: G06F 16/40 of multimedia data, e.g. slideshows comprising image and additional audio data retrieval of still image data G06F16/50; retrieval of audio data G06F16/60; retrieval of video data G06F16/70
• CPC Class Codes: G06F 16/635 Filtering based on additional data, e.g. user or group profiles
• CPC Class Codes: G06F 16/64 Browsing; Visualisation therefor generation of a list or set of audio data G06F16/638
• CPC Class Codes: G06F 16/68 Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
• CPC Class Codes: G06F 16/735 Filtering based on additional data, e.g. user or group profiles
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0201 Market modelling; Market analysis; Collecting market data
• CPC Class Codes: G06Q 30/0633 Lists, e.g. purchase orders, compilation or processing
• CPC Class Codes: Y10S 707/922 Communications
• CPC Class Codes: Y10S 707/959 Network
• CPC Class Codes: Y10S 707/99931 Database or file accessing
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching
• CPC Class Codes: Y10S 707/99935 Query augmenting and refining, e.g. inexact access
• CPC Class Codes: Y10S 707/99938 Concurrency, e.g. lock management in shared database

Abstract

A system for determining selections that a user is likely to be interested in. The determination is made based on the user'"'"'s prior indicated preferences. The user designates his or her preferred selections as entries in a user'"'"'s preference list. Entries in the user'"'"'s list are compared with entries in the other users'"'"' lists. When a significant number of matches have been found between two lists, the unmatched entries of the other user'"'"'s preference list are extracted. The unmatched entries are further processed. Those unmatched entries with a high correlation to the user'"'"'s preference list are presented to the user as selections in which the user is likely to be interested.

Claims

1. A method for outputing recommended preferences for a target user based on the target user'"'"'s input preferences and preferences of other users, the method executing on a computer system including a processor, a preferences database, an input device and an output device, wherein the preference database includes a plurality of records, one per user, where a record indicates a plurality of preferences of the user associated with that record, the method comprising the following steps:

accepting signals from the input device to generate an input record, wherein the input record indicates the target user'"'"'s preferences;

using the processor to search at least a subset of the preferences database to locate matching records containing at least one preference which is also found in the input record;

assigning each of the matching records a match count with respect to the input record, wherein a match count assigned to one record with respect to another record is equal to the number of common preferences which occur in both of the records;

identifying unmatched preferences in the matching records, where an unmatched preference is a preference found in a matching record which is not found in the input record;

assigning a weight to an unmatched preference which weight is a function of at least the frequency of occurrence of the unmatched preference in the preferences database, the weight being greater for unmatched preferences which occur less frequently;

ranking unmatched preferences by weight;

selecting recommended preferences from the unmatched preferences according to the results of the step of ranking; and

outputing, by means of the output device, an indication of the recommended preferences.

2. The method of claim 1, wherein the step of ranking unmatched preferences by weight comprises the steps of:

generating a first ranking of unmatched preferences by frequency of occurrence of the unmatched preference among all unmatched preferences;

generating a second ranking of unmatched preferences by the frequency of occurrence of the unmatched preferences among all unmatched preferences, normalized by the frequency of occurrence of the unmatched preferences among all records in the preferences database; and

generating a third ranking which is a weighted average of the first and second rankings, the third ranking being the result of the step of ranking unmatched preferences by weight.

3. The method of claim 2, further comprising the steps of:

evaluating the accuracy of the recommended preferences relative to actual preferences; and

adjusting the weights used in the weighted average to improve the accuracy of the recommended preference.

4. The method of claim 1, wherein the step of assigning a weight and the step of ranking unmatched preferences are performed for each match count in a range of interest to arrive at a ranking for each match count in the range, the method further comprising a step of calculating a weighted average of the rankings for each match count in the range.

5. The method of claim 4, further comprising the steps of:

evaluating the accuracy of the recommended preferences relative to actual preferences; and

adjusting the weights used in the weighted average of the rankings for each match count to improve the accuracy of the recommended preferences.

6. The method of claim 4, wherein the step of accepting signals is a step of accepting a fixed number, N, of target user preferences, the records also having N preferences each, and wherein the range of interest is from a match count of zero to a match count of N minus one.

7. The method of claim 4, wherein the step of accepting signals is a step of accepting ten target user preferences, the records also having ten preferences each, and wherein the range of interest is a match count of five to nine.

8. The method of claim 4, wherein calculations for which weights are set at zero weight are not performed, to save computation effort.

9. The method of claim 1, wherein the step of accepting signals from the input device is performed at a location remote from the computer system and the method further comprises a step of transmitting the input record over a data communications network to the computer system.

10. The method of claim 1, wherein the preferences are artists'"'"' names.

11. The method of claim 1, wherein the preferences are song titles.

12. The method of claim 1, wherein the preferences are movie titles.

13. An apparatus for recommending preferences comprising:

a processor;

a preferences database containing a plurality of records, each record indicating a plurality of preferences for a user associated with that record;

input means for accepting input of a target user record, the target user record being an indication of a plurality of preferences of a target user;

means for storing weighting values;

communications means for communicating data between the processor, the preferences database, the input means and the means for storing;

control means for controlling the processor to search the preferences database for records with a prespecified number of preferences in common with the target user record and to generate an initial ranking of unmatched preferences found in the located records, where an unmatched preference is a preference not found in the target user record, the initial ranking being a function of at least the frequency of occurrence of the unmatched preferences in the preferences database;

ranking means, coupled to the processor, for ranking the unmatched preferences in a subsequent ranking where the subsequent ranking is based on a function of at least the initial ranking and the weighting values; and

output means, coupled to the processor, for outputing an indication of the unmatched preferences ordered according to the subsequent ranking.

14. A method for generating a ranked list of associated objects for each object in an object database using a preferences database, the object database including a table of object records each of which references an object, and the preferences database including a table of user records each of which references a list of object preferences for a user associated with the user record, the method comprising the steps of:

(a) generating object pairs from the object database, where an object pair identifies two paired objects referenced by object records in the table of object records;

(b) for each object pair, performing the steps of;

(1) scanning the preferences database to locate user records which contain object preferences for both objects of the object pair; and

(2) calculating an object pair association weight, where the object pair association weight is a function of at least the number of user records located in step (b)(1) and is a function which results in a greater weight as more user records are located; and

(c) for each object, performing the steps of;

(1) identifying object pairs containing said object, the other object in the pair being referred to as a paired object; and

(2) ranking the paired objects from the object pairs identified in step (c)(1) where the ranking is done according to object pair association weights; and

(3) associating an ordered list of the ranked paired objects with said object as the ranked list of associated objects.

15. The method of claim 14, wherein the step of calculating an object pair association weight comprises the steps of:

identifying which object in the object pair is a less common object in the preferences database and which object is a more common object in the preferences database;

counting the number of user records located in step (b)(1) to arrive at a pair hit count;

calculating a first value as the pair hit count divided by the frequency of occurrence of the less common object in the preferences database;

calculating a second value as the pair hit count divided by the frequency of occurrence of the more common object in the preferences database;

calculating the object pair association weight as a weighted average of the first value and the second value, the relative weights being determined by a balancing value.

16. The method of claim 15, further comprising the steps of:

evaluating the accuracy of the ranked lists of associated objects; and

adjusting the balancing value to improve the accuracy of the ranked lists.

17. The method of claim 14, wherein the preferences are artist'"'"'s names.

18. The method of claim 14, wherein the preferences are song titles.

19. The method of claim 14, wherein the preferences are movie titles.

20. An apparatus for generating a ranked list of associated objects for each object in an object database, comprising:

a processor;

a preferences database, the object database including a table of object records each of which references an object and the preferences database including a table of user records each of which references a list of object preferences for a user associated with the user record;

means for generating object pairs from the object database, where an object pair identifies two paired objects referenced by object records in the table of object records;

means for scanning the preferences database to locate user records which contain object preferences for both objects of object pairs;

means for calculating an object pair association weight for each object pair, where the object pair association weight is a function of at least the number of user records containing both objects of the object pair;

means for ranking paired objects from the object pairs having a given object as one object in the object pair, where the ranking is done according to object pair association weights; and

means for outputing a ranked list of associated objects for a given object listing objects which are paired with the given object wherein the ranking in the ranked list is the ranking assigned by the means for ranking.

21. A method for determining a recommendation for a querying user based on the querying user'"'"'s preferences and the preferences of a user population, the method executing on a computer system including a preferences database, wherein the preferences database includes a plurality of user records for a plurality of users in the user population, and wherein each user record is associated with a user and contains an indication of a plurality of preferred items for that user, the method comprising the steps of:

reading, into the computer system, the indication of the plurality of preferred items of the querying user, thus forming a querying user record;

comparing the querying user record to other user records in the plurality of user records to identify user records with at least one differing item, a differing item being a preferred item which is indicated as a preferred item in the identified user record but not the querying user record;

weighting each identified user record by a weight dependent on the number of times the differing item occurs in the user records of the user population and dependent on the number of differing items in the identified user record;

ranking, by weight, the differing items according to the weights given to each identified user record containing the differing items being ranked;

presenting a ranking of the differing items to the querying user as a list of recommendations.

22. The method of claim 21, wherein the weighting step is a step of weighting each identified user record with a weight which is higher for user records with more items in common with the querying user record than for user records with less items in common with the querying record, thereby causing differing items to be ranked higher in the ranking if the differing items occur in user records with many items in common with the querying user record.

23. The method of claim 21, wherein the querying user record contains an indication of at least N preferred items and each user record contains an indication of at least N preferred items, wherein the step of weighting further comprises the steps of:

identifying, for each user record, a difference count indicating the number of differing items in the user record among the N preferred items of the user record and the N preferred items of the querying user record, the difference count ranging from zero to N;

identifying, for each differing item, a weight which is dependent on the number of times each differing item occurs in each user record for each difference count.