1. A system for large-scale, incremental clustering comprising:
a plurality of processing nodes, each comprising a processor and a non-transitory computer readable medium storing a plurality of clusters of feature vectors and machine executable instructions for determining a plurality of values for a distance metric relating each of the plurality of clusters to an input feature vector and selecting a cluster having a best value for the distance metric;
an arbitrator configured to:
receive the selected cluster and best value for the distance metric from each of the plurality of processing nodes and determine a winning cluster as one of the selected clusters and a new cluster;
accept the selected cluster if its associated best value for the distance metric meets a threshold value and to create the new cluster if the best value for the distance metric does not meet a threshold value; and
a multiplexer configured to receive the winning cluster from the arbitrator and provide the winning cluster and a new input feature vector to each of the plurality of processing nodes, wherein the winning cluster is associated with a given processing node of the plurality of processing nodes, the given processing node being configured to associate the input feature vector with the winning cluster and each other processing node of the plurality of processing nodes is instructed to delete the input feature vector.
2. The system of claim 1, a content processor configured to reduce a document to provide the input feature vector and provide the input feature vector to the multiplexer.
3. The system of claim 1, the distance metric is selectable by a user from a plurality of available distance metrics.
4. The system of claim 3, wherein the input feature vector contains nominal data and the plurality of available distance metrics include a Jaccard distance, a common element measure, a S\xf8rensen similarity index, and a Jaccard coefficient.
5. The system of claim 1, wherein the input feature vector contains ordinal data and the distance metric is a cosine distance on a unit hypersphere.
6. The system of claim 1, wherein the threshold value is selectable by a user.
7. The system of claim 1, wherein, when the winning cluster is determined to be the new cluster, the arbitrator is configured to randomly assign the new cluster to one of the plurality of processing nodes.
8. The system of claim 1, wherein, when the winning cluster is determined to be the new cluster, the arbitrator is configured to assign the new cluster to one of the plurality of processing nodes according to a load balancing algorithm.
9. The system of claim 1, wherein the plurality of clusters stored at each processing node each have an associated representative vector, the processing nodes determining the plurality of values for the distance metric as to relate the input feature vector to each of the representative vectors.
10. The system of claim 9, wherein the representative vector for each of the plurality of clusters is a feature vector within the cluster with the smallest distance sum to all other feature vectors in the cluster.
11. The system of claim 9, wherein the representative vector for each of the plurality of clusters is determined according to one of a plurality of centroid assignment methods, and the user selects the one of the plurality of centroid assignment methods.
12. The system of claim 9, wherein the representative vector for each of the plurality of clusters is a first vector placed within each cluster updated by shorter length vectors until a minimum-specified-length vector is achieved.
13. A method for large-scale, incremental clustering comprising:
distributing an input feature vector to each of a plurality of processing nodes from a multiplexer, each processing node comprising a processor and a non-transitory computer readable medium;
determining, at each of the plurality of processing nodes, a plurality of values for a distance metric, with each value representing the similarity of the input feature vector to a cluster stored at the processing node;
selecting a best cluster at each of the plurality of processing nodes according to the plurality of values for the distance metric determined at the processing node;
selecting an overall best cluster from the selected best clusters according to their associated values of the distance metric;
selecting the overall best cluster as a winning cluster if the distance metric associated with the overall best cluster meets a threshold value;
creating a new cluster as the winning cluster if the distance metric associated with the overall best cluster does not meet a threshold value;
providing the identity of the winning cluster from an arbitrator to the multiplexer; and
distributing the identity of the winning cluster associated with the input feature vector to the plurality of processing nodes along with a new input feature vector for analysis, wherein the winning cluster is associated with a given processing node of the plurality of processing nodes, the given processing node being configured to associate the input feature vector with the winning cluster and each other processing node of the plurality of processing nodes is instructed to delete the input feature vector.
14. The method of claim 13, further comprising allowing the user to select the distance metric from a plurality of available distance metrics.
15. The method of claim 13, further comprising allowing the user to select the threshold value.
16. A large-scale document clustering system comprising:
a content processor configured to reduce a document to provide an input feature vector;
a multiplexer configured to distribute the input feature vector to a plurality of processing nodes;
the plurality of processing nodes, each comprising a processor and a non-transitory computer readable medium storing a plurality of clusters of feature vectors and machine executable instructions for determining a plurality of values for a distance metric relating each of the plurality of clusters to the input feature vector and selecting a cluster having a best value for the distance metric; and
an arbitrator configured to:
receive the selected cluster and best value for the distance metric from each of the plurality of processing nodes and determine a winning cluster as one of the selected clusters and a new cluster; and
accept the selected cluster if its associated best value for the distance metric meets a threshold value and to create the new cluster if the best value for the distance metric does not meet a threshold value,
wherein the multiplexer is configured to receive the winning cluster from the arbitrator and provide the winning cluster when a next input feature vector is distributed to the plurality of processing nodes, the winning cluster being associated with a given processing node of the plurality of processing nodes, the given processing node being configured to associate the input feature vector with the winning cluster and each other processing node of the plurality of processing nodes is instructed to delete the input feature vector.
17. The system of claim 16, the system further comprising a dictionary having entries for each of a plurality of terms, the content processor generating the input feature vector as a series of values representing dictionary entries for terms located in the document, such that the input feature vector communicates the presence or absence of each term in the document.
The claims below are in addition to those above.
All refrences to claim(s) which appear below refer to the numbering after this setence.
1. A method for distributing advertisement information images, the method comprising:
loading an advertisement information image into a subscriber’s cellular phone;
demonstrating the advertisement information image at the cellular phone;
viewing the advertisement information image; and
compensating the user’s expenses arisen from using the cellular phone; characterized in that the subscriber fills up a questionnaire, on which basis the advertiser selects subscribers for distributing hisher advertisement information image, as which to the cellular phone subscriber is sent the advertisement information image for which the subscriber has subscribed when filling up the questionnaire; wherein the step of loading said advertisement information image to the cellular phone subscriber being carried out with hisher consent and according to the media plan of the advertiser till receiving to the cellular phone an incomingoutgoing call, or incomingoutgoing SMS andor MMS message; the step of demonstrating said advertisement information image being carried out at the time of the incomingoutgoing call or incomingoutgoing SMS andor MMS message; the step of viewing said advertisement information image being carried out by the subscriber till the time of the call reply, andor at the time of the call termination, andor till the time of the SMS and MMS message view and reply, the advertisement information image being on the mobile phone screen during entire continuance of the conversation or until the call release andor the SMS and MMS message reply, and the step of compensating the user’s expenses being carried out by means of receiving monetary reward to the user’s account for every demonstration of the advertisement information image.
2. The method according to claim 1, characterized in that when receiving a new SMS andor MMS message at the time of viewing the advertisement information image, on the phone screen emerges a notification not impeding to view the advertisement information and meaning the fact of receiving the new SMS or MMS message for the subscriber or a new advertisement information image.
3. The method according to claim 1, characterized in that the advertisement information image presented on the mobile phone screen is accompanied with a corresponding voice or musical or sound accompaniment being a part of the advertisement information image.
4. The method according to claim 1, characterized in that the advertisement information image coming to the subscriber’s mobile phone screen is scaled in accordance with the size of the mobile phone screen.
5. The method according to claim 1, characterized in that the monetary reward received to the subscriber’s account could be used by himher for paying the communication service andor any other services, andor cashed by the subscriber.