Machine learning system

1. A method for training a classifier to classify elements of a data set according to a predetermined characteristic, said data set including N elements, said elements each characterized by at least one feature, said method comprising:

forming a first labeled subset of elements from said data set, said elements of said first labeled subset each labeled according to whether said element includes said predetermined characteristic;

training an algorithmic classifier to classify for said predetermined characteristic according to said first labeled subset thereby determining which at least one feature is relevant to classifying for said predetermined characteristic; and

querying with said classifier an inverted index, said inverted index formed over said at least one feature and generated from said data set, thereby generating a ranked set of elements from said data set,

wherein said method further comprises:

forming a second labeled subset, said second labeled subset including a predetermined number P of elements labeled according to whether each element includes said characteristic, wherein said P elements in said second labeled subset includes a first subset of K=α×P elements selected randomly from said N elements of said data set, with number α ranging from 0 to 1, and a second subset of M=(1−α)×P elements selected from said ranked set of elements.

2. A method for training a classifier according to claim 1 , wherein said method further comprises:

weighting each element in said first subset of said second labeled subset with a first weighting parameter ω α ; and

weighting each element in said second subset of said second labeled subset with a second weighting parameter ω M .

3. A method for training a classifier according to claim 2 , wherein said first weighting parameter ω α =1/α.

4. A method for training a classifier according to claim 2 , wherein said second weighting parameter ω M =M/N.

5. A method for training a classifier according to claim 1 , wherein said method further comprises:

forming a new labeled subset including said first and second labeled subsets; and

further training said algorithmic classifier on said new labeled subset.

6. A method for training a classifier to classify a sequence of elements of a data set according to a predetermined characteristic, said elements characterized by at least one feature, said method comprises:

forming a first labeled subset of element sequences by assigning a label to each element in said element sequence indicating whether said element is a member of a sequence of elements in said element sequence that includes said predetermined characteristic;

training an algorithmic classifier to predict the label assigned to each element in said element sequence according to said first labeled subset thereby determining which at least one feature is relevant to classifying whether said element sequence includes said sequence of elements that includes said predetermined characteristic; and

querying with said classifier an inverted index, said inverted index formed over said at least one feature and generated from said data set, thereby generating a ranked set of element sequences from said data set,

wherein said classifier further bases its prediction of said label assigned to each element on the label assigned to at least one element prior to said each element in said element sequence.

7. A method for training a classifier to classify a sequence of elements of a data set according to claim 6 , further comprising assigning a probability associated with each type of label transition for each element in said element sequence thereby determining which at least one feature is relevant to classifying for said each type of label transition and hence a boundary of said sequence of elements within said element sequence that includes aid predetermined characteristic.

8. A method for training a classifier according to claim 6 , wherein said method further comprises:

forming a second labeled subset, said second labeled subset including a predetermined number P of element sequences, wherein each element of said P element sequences is labeled according to whether each element is a member of a sequence of elements that includes said predetermined characteristic and wherein said P element sequences are selected from said ranked set of element sequences.

9. A method for training a classifier according to claim 8 , wherein said method further comprises:

forming a new labeled subset including said first and second labeled subsets; and further training said algorithmic classifier on said new labeled subset.

10. A classification system for classifying elements of a data set according to a predetermined characteristic, said data set including N elements, said elements each characterized by at least one feature, the system comprising:

a database for storing said elements;

a labeler to label a first labeled subset of elements from said data set, said elements of said first labeled subset each labeled according to whether said element includes said predetermined characteristic;

an algorithmic classifier training to classify for said predetermined characteristic according to said first labeled subset thereby determining which at least one feature is relevant to classifying for said predetermined characteristic;

an indexer to form an inverted index generated from said data elements said inverted index formed over said at least one feature; and

a ranker to form a ranked set of data elements, wherein said ranked set is formed by querying with said classifier said inverted index,

wherein said system further comprises:

said labeler forming a second labeled subset, said second labeled subset including a predetermined number P of elements labeled according to whether each element includes said characteristic, wherein said P elements in said second labeled subset includes a first subset of K=α×P elements selected randomly from said N elements of said data set, with number a ranging from 0 to 1, and a second subset of M=(1-α)×P elements selected from said ranked set of elements.

11. A classification system for classifying elements of a data set according to claim 10 , wherein said system further comprises:

weighting each element in said first subset of said second labeled subset with a first weighting parameter ω α ; and

weighting each element in said second subset of said second labeled subset with a second weighting parameter ω M .

12. A classification system for classifying elements of a data set according to claim 11 , wherein said first weighting parameter ωα=1/α.

13. A classification system for classifying elements of a data set according to claim 11 , wherein said second weighting parameter ω αM =M/N.

14. A classification system for classifying elements of a data set according to claim 10 , wherein said system further comprises:

said labeler forming a new labeled subset including said first and second labeled subsets; and

said algorithmic classifier training on said new labeled subset.

15. A classification system for classifying a sequence of elements of a data set according to a predetermined characteristic, said elements characterized by at least one feature, said system comprising:

a database for storing said elements;

a labeler to label a first labeled subset of element sequences by assigning a label to each element in said element sequence indicating whether said element is a member of a sequence of elements in said element sequence that includes said predetermined characteristic;

an algorithmic classifier trained to predict the label assigned to each element in said element sequence according to said first labeled subset thereby determining which at least one feature is relevant to classifying whether said element sequence includes said sequence of elements that includes said predetermined characteristic;

an indexer to form an inverted index generated from said data elements, said inverted index formed over said at least one feature; and

a ranker to form a ranked set of element sequences, wherein said ranked set is formed by querying with said classifier said inverted index,

wherein said classifier further bases its prediction of said label assigned to each element on the label assigned to at least one element prior to said each element in said element sequence.

16. A classification system for classifying a sequence of elements of a data set according to claim 15 , further comprising assigning a probability associated with each type of label transition for each element in said element sequence thereby determining which at least one feature is relevant to classifying for said each type of label transition by said classifier and hence a boundary of said sequence of elements within said element sequence that comprises said predetermined characteristic.

17. A classification system for classifying a sequence of elements of a data set according to claim 15 , wherein said system further comprises:

said labeler forming a second labeled subset, said second labeled subset including a predetermined number P of element sequences, wherein each element of said P element sequences is labeled according to whether each element is a member of a sequence of elements that includes said predetermined characteristic and wherein said P element sequences are selected from said ranked set of element sequences.

18. A classification system for classifying a sequence of elements of a data set according to claim 17 , wherein said system further comprises:

said labeler forming a new labeled subset including said first and second labeled subsets; and said algorithmic classifier training on said new labeled subset.