System and Method for Generating Queries

• US 20100211569A1
• Filed: 02/18/2009
• Published: 08/19/2010
• Est. Priority Date: 02/18/2009
• Status: Active Grant

• Current Assignee: Avaya Incorporated
• Original Assignee: Avaya Incorporated
• Inventors: Ozgur, Arzucan, Skiba, David, Erhart, George, Matula, Valentine
• Granted Patent: US 8,301,619 B2
• Time in Patent Office: Days
• US Class Current: 707/737
• CPC Class Codes: G01R 31/52 Testing for short-circuits, leakage current or ground faults
• CPC Class Codes: G01R 31/54 Testing for continuity
• CPC Class Codes: G06F 16/3322 using system suggestions G06F16/3325 takes precedence

Abstract

The system utilizes (gets) training data that comprises a plurality of training documents. Each of the plurality of training documents comprises a training token(s). The plurality of training documents are clustered into a plurality of clusters based on at least one training token in the plurality of training documents. Each cluster contains at least one training document. A Boolean query(s) is generated for a cluster based on an occurrence of the at least one training token in a training document in the plurality of training documents. The system gets production data that comprises a plurality of production documents. Each of the plurality of production documents comprises a production token(s). The Boolean query(s) is then executed on the production data.

Claims

1. A system for generating a Boolean query comprising:

a. a data manager configured to get training data and production data, wherein the training data comprises a plurality of training documents and each of the plurality of training documents comprises at least one training token, and wherein the production data comprises a plurality of production documents and each of the plurality of production documents comprises at least one production token;

b. a clustering manager configured to cluster the plurality of training documents into a plurality of clusters based on at least one training token in the plurality of training documents, wherein each cluster comprises at least one training document; and

c. a query manager configured to generate the Boolean query for a cluster of the plurality of clusters based on an occurrence of the at least one training token in at least one training document of the plurality of training documents, and to execute the Boolean query on the plurality of production documents in the production data.

2. The system of claim 1, wherein the at least one training token comprises a plurality of training tokens and wherein the data manager is configured to calculate a training token/cluster weight matrix for the plurality of training tokens for the cluster, to rank the plurality of training tokens for the cluster according to the matrix, and to select a list of top N training tokens from the ranked plurality of training tokens for the cluster, where N is a positive integer.

3. The system of claim 2, wherein the query manager is configured to set a clause equal to an empty clause, select a next best training token from the top N training tokens, determine if a recall improvement is greater than a minimum improvement, determine if a number of clauses is less than a maximum number of clauses in response to the recall improvement not being greater than the minimum improvement, to respond to determining that the recall improvement is not greater than the minimum improvement by setting the clause to the selected next best training token, to respond to the recall improvement being greater than the minimum improvement by selecting the next best training token, determine if a precision improvement is greater than a recall drop, set a new value of a query equal to a current value of the query ORed with the clause responsive to the precision improvement not being greater than the recall drop, set a new value of the clause equal to a current value of the clause ANDed with the selected next best training token responsive to the precision improvement being greater than the recall drop, determine if a number of clause tokens is less than a maximum number of clause tokens, select the next best training token responsive to the number of clause tokens being less than the maximum number of clause tokens, set the query equal to the query ORed with the clause responsive to the number of clause tokens not being less than the maximum number of clause tokens, determine if the number of clauses is less than the maximum number of clauses, set the clause to the empty clause responsive to the number of clauses being less than the maximum number of clauses, and execute the query on the production data responsive to the number of clauses not being less than the maximum number of clauses.

4. The system of claim 2, wherein the query manager is configured to select a next best positive training token from the top N training tokens, determine if a recall improvement is greater than a minimum improvement, determine if a number of OR tokens is less than a maximum number of OR tokens responsive to the recall improvement not being greater then the minimum improvement, to respond to the recall improvement being greater than the minimum improvement by setting a new value of a query equal to a current value of the query ORed with the selected next best positive training token, to respond to the recall improvement not being greater than the minimum improvement by determining if the number of OR tokens is less than the maximum number of OR tokens, select the next best positive training token responsive to the number of OR tokens being less than the maximum number of OR tokens, select the next best negative training token responsive to the number of OR tokens not being less than the maximum number of OR tokens, determine if a precision improvement is greater than a recall drop, respond to the precision improvement not being greater than the recall drop by determining if a number of AND NOT tokens is less than a maximum number of AND NOT token, set a new value of the query equal to a current value of the query ANDed with the selected next best negative training token responsive to the precision improvement being greater than the recall drop, determine if the number of AND NOT tokens is less than the maximum number of AND NOT tokens, select the next best negative training token responsive to the number of AND NOT tokens being less than the maximum number of AND NOT tokens, and execute the Boolean query on the production data responsive to the number of AND NOT tokens not being less than the maximum number of AND NOT tokens.

5. The system of claim 2, wherein the query manager is configured to generate the Boolean query by an algorithm selected from the group comprising:

the algorithm defined in FIG. 10 and the algorithm defined in FIG. 11.

6. The system of claim 1, wherein the data manager is configured to analyze results from the Boolean query, determine if there is new training data, get new training data responsive to there being new training data, determine if the Boolean query has been modified responsive to there not being new training data, execute the Boolean query on the plurality of production documents in the production data responsive to the Boolean query being modified, determine if there is a change in the number of clusters responsive to the Boolean query not being modified, cluster the plurality of training documents into a plurality of clusters based on at least one training token in the plurality of training documents responsive to there being a change in the number of clusters, respond to there not being a change in the number of clusters by determining if there is new production data, get production data responsive to either (a) there being new production data or (b) there being new production data and the Boolean query having been modified, being done responsive to either (c) there not being new production data or (d) there not being new production data and the Boolean query having not been modified.

7. The system of claim 1, wherein the data manager is configured to get second production data, wherein the second production data comprise a plurality of second production documents and wherein each of the plurality of second production documents comprises at least one second production token and the query manager is configured to execute the Boolean query on the plurality of second production documents in the second production data.

8. The system of claim 1, wherein clustering the training data is accomplished by an algorithm selected from the group comprising:

a k-means, a bisecting k-means, an agglomerative, and a divisive hierarchical.

9. The system of claim 1, wherein the plurality of training documents and/or the plurality of production documents comprises at least one item selected from the group comprising:

a document, an email, a content of a website, a content of a blog, a transcript of a voice call, an identifiable item from a video, an identifiable item from a picture, a transcript of a conversation, a document containing different languages, and a transcript of an audio portion of a video.

10. The system of claim 1, wherein the training token is an item selected from the group comprising:

a word, a part of a word, a phrase, a bi-gram, a tri-gram, and an N-gram.

11. A system for generating a Boolean query comprising:

a. a data manager configured to get training data and production data, wherein the training data comprises a plurality of training documents and each of the plurality of training documents comprises at least one training token, wherein the production data comprises a plurality of production documents and each of the plurality of production documents comprises at least one production token, and wherein data manager is configured to clean the training data, and identify at least one salient token from the at least one training token in each of the plurality of training documents;

b. a clustering manager configured to cluster the plurality of training documents into a plurality of clusters based on at least one training token in the plurality of training documents or the at least one salient token, wherein each cluster comprises at least one training document; and

c. a query manager configured to generate the Boolean query for a cluster of the plurality of clusters based on an occurrence of the at least one salient token in at least one training document of the plurality of training documents, and to execute the Boolean query on the plurality of production documents in the production data.

12. The system of claim 11, wherein the at least one salient token comprises a plurality of salient tokens and wherein the data manager is further configured to calculate a salient token/cluster weight matrix for the plurality of salient tokens for the cluster, to rank the plurality of salient tokens for the cluster according to the matrix, and to select a list of the top N salient tokens from the ranked plurality of training tokens for the cluster, where N is a positive integer.

13. The system of claim 12, wherein the query manager is configured to set a clause equal to an empty clause, select a next best salient token from the top N salient tokens, determine if a recall improvement is greater than a minimum improvement, determine if a number of clauses is less than a maximum number of clauses responsive to the recall improvement not being greater than the minimum improvement, to respond to determining that the recall improvement is not greater than the minimum improvement by setting the clause to the selected next best salient token, to respond to the recall improvement being greater than the minimum improvement by selecting the next best salient token, determine if a precision improvement is greater than a recall drop, set new value of a query equal to a current value of the query ORed with the clause responsive to the precision improvement not being greater than the recall drop, set a new value of the clause equal to a current value of the clause ANDed with the selected next best salient token responsive to the precision improvement being greater than the recall drop, determine if the number of clause tokens is less than the maximum number of clause tokens, select the next best salient token responsive to the number of clause tokens being less than the maximum number of clause tokens, set the query equal to the query ORed with the clause responsive to the number of clause tokens not being less than the maximum number of clause tokens, determine if a number of clauses is less than a maximum number of clauses, set the clause to the empty clause responsive to the number of clauses being less than the maximum number of clauses, and execute the Boolean query on the production data responsive to the number of clauses not being less than the maxim number of clauses.

14. The system of claim 12, wherein the query manager is configured to select a next best positive salient token from the top N salient tokens, determine if a recall improvement is greater than a minimum improvement, determine if a number of OR tokens is less than a maximum number of OR tokens responsive to the recall improvement not being greater than the minimum improvement, to respond to the recall improvement being greater than the minimum improvement by setting a new value of a query equal to a current value of the query ORed with the selected next best positive salient token, to respond to the recall improvement not being greater than the minimum improvement by determining if the number of OR tokens is less than the maximum number of OR tokens, select the next best positive salient token responsive to the number of OR tokens being less than the maximum number of OR tokens, select the next best negative salient token responsive to the number of OR tokens not being less than the maximum number of OR tokens, determine if a precision improvement is greater than a recall drop, respond to the precision improvement not greater than the recall drop by determining if a number of AND NOT tokens is less than a maximum number of AND NOT tokens, set a new value of the query equal to a current value of the query ANDed with the selected next best negative salient token responsive to the precision improvement being greater than the recall drop, determine if the number of AND NOT tokens is less than the maximum number of AND NOT tokens, select the next best negative salient token responsive to the number of AND NOT tokens being less than the maximum number of AND NOT tokens, and execute the Boolean query on the production data responsive to the number of AND NOT tokens not being less than the maximum number of AND NOT tokens.

15. The system of claim 12, wherein the query manager is configured to generate the Boolean query by an algorithm selected from the group comprising:

the algorithm defined in FIG. 10 and the algorithm defined in FIG. 11.

16. The system of claim 11, wherein the data manager is configured to analyze results from the Boolean query, determine if there is new training data, get new training data responsive to there being new training data, determine if the Boolean query has been modified responsive to there not being new training data, execute the Boolean query on the plurality of production documents in the production data responsive to the Boolean query being modified, determine if there is a change in the number of clusters responsive to the Boolean query not being modified, cluster the plurality of training documents into a plurality of clusters based on at least one training token in the plurality of training documents responsive to there being a change in the number of clusters, respond to there not being a change in the number of clusters by determining if there is new production data, get production data responsive to either (a) there being new production data or (b) there being new production data and the Boolean query having been modified, being done responsive to either (c) there not being new production data or (d) there not being new production data and the Boolean query having not been modified.

17. The system of claim 11, wherein the data manager is configured to get a second production data, wherein the second production data comprises a plurality of second production documents and wherein each of the plurality of second production documents comprises a second production token and the query manager is configured to execute the Boolean query on the plurality of second production documents in the second production data.

18. The system of claim 11, wherein identifying the at least one salient token from the at least one training token in each of the plurality of training documents is accomplished by an algorithm selected from the group comprising:

a Chi-square test, likelihood ratios, mutual information, and a t-test.

19. The system of claim 11, wherein clustering the training data is accomplished by an algorithm selected from the group comprising:

a k-means, a bisecting k-means, an agglomerative, and a divisive hierarchical.

20. The system of claim 11, wherein the plurality of training documents and/or the plurality of production documents comprises at least one item selected from the group comprising:

a document, an email, a content of a website, a content of a blog, a transcript of a voice call, an identifiable item from a video, an identifiable item from a picture, a transcript of a conversation, a document containing different languages, and a transcript of an audio portion of a video.

21. The system of claim 11, wherein the salient token is an item selected from the group comprising:

a word, a part of a word, a phrase, a bi-gram, a tri-gram, and an N-gram.

22. A computer implemented method for generating a Boolean query comprising:

a. getting training data, wherein the training data comprises a plurality of training documents and wherein each of the plurality of training documents comprises at least one training token;

b. clustering the plurality of training documents into a plurality of clusters based on at least one training token of the plurality of training documents, wherein each cluster comprises at least one training document;

c. generating the Boolean query for a cluster of the plurality of clusters based on an occurrence of the at least one training token in at least one training document of the plurality of training documents;

d. getting production data, wherein the production data comprises a plurality of production documents and wherein each of the plurality of production documents comprises at least one production token; and

e. executing the Boolean query on the plurality of production documents in the production data.

23. The method of claim 22, wherein the at least one training token comprises a plurality of training tokens and wherein the step of generating the Boolean query for the cluster in the plurality of clusters further comprises the steps of:

f. calculating a training token/cluster weight matrix for the plurality of training tokens for the cluster;

g. ranking the plurality of training tokens for the cluster according to the matrix; and

h. selecting a list of the top N training tokens from the ranked plurality of training tokens for the cluster, where N is a positive integer.

24. The method of claim 23, wherein generating the Boolean query further comprising the steps of:

i. setting a clause to an empty clause;

j. selecting a next best training token from the top N training tokens;

k. determining if a recall improvement is greater than a minimum improvement;

l. responsive to the recall improvement not being greater than the minimum improvement, going to step (u);

m. responsive to the recall improvement being greater than the minimum improvement, setting the clause equal to the selected next best training token;

n. selecting the next best training token from the top N training tokens;

o. determining if a precision improvement is greater than a recall drop;

p. responsive to the precision improvement not being greater than the recall drop, going to step (t);

q. responsive to the precision improvement being greater than the recall drop, setting the clause equal to the clause AN Ded with the selected next best training token;

r. determining if a number of clause tokens is less than a maximum number of clause tokens;

s. responsive to the number of clause tokens being less than the maximum number of clause tokens, going to step (n);

t. responsive to the number of clause tokens not being less than the maximum number of clause tokens, setting the query equal to the query ORed with the clause;

u. determining if a number of clauses is less than a maximum number of clauses;

v. responsive to the number of clauses being less than the maximum number of clauses, going to step (i); and

w. responsive to the number of clauses not being less than the maximum number of clauses, going to step (d).

25. The method of claim 23, wherein generating the Boolean query further comprises the steps of:

i. selecting a next best positive training token from the top N training tokens;

j. determining if a recall improvement is greater than a minimum improvement;

k. responsive to the recall improvement not being greater than the minimum improvement, going to step (m);

l. responsive to the recall improvement being greater than the minimum improvement, setting a query equal to the query ORed with the selected next best positive training token;

m. determining if a number of OR tokens is less than a maximum number of OR tokens;

n. responsive to the number of OR tokens being less than the maximum number of OR tokens, going to step (i);

o. responsive to the number of OR tokens not being less than the maximum number of OR tokens, selecting a next best negative training token from the top N training tokens;

p. determining if a precision improvement is greater than a recall drop;

q. responsive to the precision improvement not being greater than the recall drop, going to step (s);

r. responsive to the precision improvement being greater than the recall drop, setting the query equal to the query ANDed with the selected next best negative training token;

s. determining if a number of AND NOT tokens is less than a maximum number of AND NOT tokens;

t. responsive to the number of AND NOT tokens being less than the maximum number of AND NOT tokens, going to step (o); and

u. responsive to the number of AND NOT tokens not being less than the maximum number of AND NOT tokens, going to step (d).

26. The method of claim 23, wherein generating the Boolean query is accomplished by an algorithm selected from the group comprising:

the algorithm defined in FIG. 10 and the algorithm defined in FIG. 11.

27. The method of claim 22, further comprising the steps of:

f. analyzing results from the Boolean query;

g. determining if there is new training data;

h. responsive to there being new training data, going to step (a);

i. responsive to there not being new training data, determining if the Boolean query has been modified;

j. responsive to the Boolean query being modified, going to step (e);

k. responsive to the Boolean query not being a modified, determining if there is a change in the number of clusters;

l. responsive to there being a change in the number of clusters, going to step (b);

m. responsive to there not being a change in the number of clusters, determining if there is new production data, or if there is new production data and the Boolean query has been modified;

n. responsive to there being new production data, or there being new production data and the Boolean query has been modified, going to step (d); and

o. responsive to there not being new production data or there not being new production data and the Boolean query has not been modified, being done.

28. The method of claim 22, further comprising the steps of:

f. getting a second production data, wherein the second production data comprises a plurality of second production documents and wherein each of the plurality of second production documents comprises a second production token; and

g. executing the Boolean query on the plurality of second production documents in the second production data.

29. The method of claim 22, wherein clustering the training data is accomplished by an algorithm selected from the group comprising:

a k-means, a bisecting k-means, an agglomerative, and a divisive hierarchical.

30. The method of claim 22, wherein the plurality of training documents and/or the plurality of production documents comprises at least one item selected from the group comprising:

a document, an email, a content of a website, a content of a blog, a transcript of a voice call, an identifiable item from a video, an identifiable item from a picture, a transcript of a conversation, a document containing different languages, and a transcript of an audio portion of a video.

31. The method of claim 22, wherein the at least one training token is an item selected from the group comprising:

a word, a part of a word, a phrase, a bi-gram, a tri-gram, and an N-gram.

43. An apparatus for performing the method of one of claims 22-42

32. A computer implemented method for generating a Boolean query comprising:

a. getting training data, wherein the training data comprises a plurality of training documents and wherein each of the plurality of training documents comprises at least one training token;

b. cleaning the training data;

c. identifying at least one salient token from the at least one training token in each of the plurality of training documents;

d. clustering the plurality of training documents into a plurality of clusters based on at least one training token in the plurality of training documents or the at least one salient token, wherein each cluster comprises at least one training document;

e. generating the Boolean query for a cluster of the plurality of clusters based on an occurrence of at least one salient token in at least one training document of the plurality of training documents;

f. getting production data, wherein the production data comprises a plurality of production documents and wherein each of the plurality of production documents comprises at least one production token; and

g. executing the Boolean query on the plurality of production documents in the production data.

33. The method of claim 32, wherein the at least one salient token comprises a plurality of salient tokens and wherein the step of generating the Boolean query for the cluster in the plurality of clusters further comprises the steps of:

h. calculating a salient token/cluster weight matrix for the plurality of salient tokens for the cluster;

i. ranking the plurality of salient tokens for the cluster according to the matrix; and

j. selecting a list of the top N salient tokens from the ranked plurality of salient tokens for the cluster, where N is a positive integer.

34. The method of claim 33, wherein generating the Boolean query further comprising the steps of:

k. setting a clause to an empty clause;

l. selecting a next best salient token from the top N salient tokens;

m. determining if a recall improvement is greater than a minimum improvement;

n. responsive to the recall improvement not being greater than the minimum improvement, going to step (w);

o. responsive to the recall improvement being greater than the minimum improvement, setting the clause equal to the selected next best salient token;

p. selecting the next best salient token from the top N salient tokens;

q. determining if a precision improvement is greater than a recall drop;

r. responsive to the precision improvement not being greater than the recall drop, going to step (v);

s. responsive to the precision improvement being greater than the recall drop, setting the clause to the clause ANDed with the selected next best salient token;

t. determining if a number of clause tokens is less than a maximum number of clause tokens;

u. responsive to the number of clause tokens being less than the maximum number of clause tokens, going to step (p);

v. responsive to the number of clause tokens not being less than the maximum number of clause tokens, setting the query equal to the query ORed with the clause;

w. determining if a number of clauses is less than a maximum number of clauses;

x. responsive to the number of clauses being less than the maximum number of clauses, going to step (k); and

y. responsive to the number of clauses not being less than the maximum number of clauses, going to step (f).

35. The method of claim 33, wherein generating the Boolean query further comprises the steps of:

k. selecting a next best positive salient token from the top N salient tokens;

l. determining if a recall improvement is greater than a minimum improvement;

m. responsive to the recall improvement not being greater than the minimum improvement, going to step (o);

n. responsive to the recall improvement being greater than the minimum improvement, setting a query equal to the query ORed with the selected next best positive salient token;

o. determining if a number of OR tokens is less than a maximum number of OR tokens;

p. responsive to the number of OR tokens being less than the maximum number of OR tokens, going to step (k);

q. responsive to the number of OR tokens not being less than the maximum number of OR tokens, selecting a next best negative salient token from the top N salient tokens;

r. determining if a precision improvement is greater than a recall drop;

s. responsive to the precision improvement not being greater than the recall drop, going to step (u);

t. responsive to the precision improvement being greater than the recall drop, setting the query equal to the query ANDed with the selected next best negative salient token;

u. determining if a number of AND NOT tokens is less than a maximum number of AND NOT tokens;

v. responsive to the number of AND NOT tokens being less than the maximum number of AND NOT tokens, going to step (q); and

w. responsive to the number of AND NOT tokens not being less than the maximum number of AND NOT tokens, going to step (f).

36. The method of claim 33, wherein generating the Boolean query is accomplished by an algorithm selected from the group comprising:

the algorithm defined in FIG. 10 and the algorithm defined in FIG. 11.

37. The method of claim 32, further comprising the steps of:

h. analyzing results from the Boolean query;

i. determining if there is new training data;

j. responsive to there being new training data, going to step (a);

k. responsive to there not being new training data, determining if the Boolean query has been modified;

l. responsive to the Boolean query being modified, going to step (g);

m. responsive to the Boolean query not being a modified, determining if there is a change in the number of clusters;

n. responsive to there being a change in the number of clusters, going to step (d);

o. responsive to there not being a change in the number of clusters, determining if there is new production data, or if there is new production data and the Boolean query has been modified;

p. responsive to there being new production data, or there being new production data and the Boolean query has been modified, going to step (f); and

q. responsive to there not being new production data or there not being new production data and the Boolean query has not been modified, being done.

38. The method of claim 32, further comprising the steps of:

h. getting a second production data, wherein the second production data comprises a plurality of second production documents and wherein each of the plurality of second production documents comprises a second production token; and

i. executing the Boolean query on the plurality of second production documents in the second production data.

39. The method of claim 32, wherein identifying the at least one salient token from the at least one training token in each of the plurality of training documents is accomplished by an algorithm selected from the group comprising:

a Chi-square test, likelihood ratios, mutual information, and a t-test.

40. The method of claim 32, wherein clustering the training data is accomplished by an algorithm selected from the group comprising:

a k-means, a bisecting k-means, an agglomerative, and a divisive hierarchical.

41. The method of claim 32, wherein the plurality of training documents and/or the plurality of production documents comprises at least one item selected from the group comprising:

a document, an email, a content of a website, a content of a blog, a transcript of a voice call, an identifiable item from a video, an identifiable item from a picture, a transcript of a conversation, a document containing different languages, and a transcript of an audio portion of a video.

42. The method of claim 32, wherein the salient token is an item selected from the group comprising:

a word, a part of a word, a phrase, a bi-gram, a tri-gram, and an N-gram.

44. An apparatus for generating a Boolean query comprising:

a. means for getting training data, wherein the training data comprises a plurality of training documents and wherein each of the plurality of training documents comprises at least one training token;

b. means for clustering the plurality of training documents into a plurality of clusters based on at least one training token in the plurality of training documents, wherein each cluster comprises at least one training document;

c. means for generating the Boolean query for a cluster in the plurality of clusters based on an occurrence of the at least one training token in at least one training document in the plurality of training documents;

d. means for getting production data, wherein the production data comprises a plurality of production documents and wherein each of the plurality of production documents comprises at least one production token; and

e. means for executing the Boolean query on the plurality of production documents in the production data.

45. An apparatus for generating a Boolean query:

a. means for getting training data, wherein the training data comprises a plurality of training documents and wherein each of the plurality of training documents comprises at least one training token;

b. means for cleaning the training data;

c. means for identifying at least one salient token from the at least one training token in each of the plurality of training documents;

d. means for clustering the plurality of training documents into a plurality of clusters based on at least one training token in the plurality of training documents or the at least one salient token, wherein each cluster comprises at least one training document;

e. means for generating the Boolean query for a cluster in the plurality of clusters based on an occurrence of the at least one salient token in at least one training document in the plurality of training documents;

f. means for getting production data, wherein the production data comprises a plurality of production documents and wherein each of the plurality of production documents comprises at least one production token; and

g. means for executing the Boolean query on the plurality of production documents in the production data.

46. A system for generating a Boolean query comprising:

a. a data manager configured to get training data and production data, wherein the training data comprises a plurality of training documents and each of the plurality of training documents comprises at least one training token, wherein the production data comprises a plurality of production documents and each of the plurality of production documents comprises at least one production token, and wherein the data manager is configured to clean the training data, identify a plurality of salient tokens from the at least one training token in each of the plurality of training documents, calculate a salient token/cluster weight matrix for the plurality of salient tokens for the cluster, to rank the plurality of salient tokens for the cluster, and to select a list of the top N salient tokens from the plurality of training tokens for the cluster;

b. a clustering manager configured to cluster the plurality of training documents into a plurality of clusters based on at least one training token in the plurality of training documents, wherein each cluster comprises at least one training document; and

c. a query manager configured to generate the Boolean query for a cluster in the plurality of clusters based on an occurrence of the plurality of salient tokens in at least one training document in the plurality of training documents, and to execute the Boolean query on the plurality of production documents in the production data.

47. A computer implemented method for generating a Boolean query comprising:

a. getting training data, wherein the training data comprises a plurality of training documents and wherein each of the plurality of training documents comprises at least one training token;

b. cleaning the training data;

c. identifying a plurality of salient tokens from the at least one training token in each of the plurality of training documents;

d. clustering the plurality of training documents into a plurality of clusters based on at least one training token in the plurality of training documents, wherein each cluster comprises at least one training document;

e. calculating a salient token/cluster weight matrix for the plurality of salient tokens for the cluster;

k. ranking the plurality of salient tokens for the cluster; and

l. selecting a list of the top N salient tokens from the plurality of salient tokens for the cluster;

m. generating the Boolean query for a cluster of the plurality of clusters based on an occurrence of the plurality of salient tokens in the at least one training document in the plurality of training documents;

n. getting production data, wherein the production data comprises a plurality of production documents and wherein each of the plurality of production documents comprises at least one production token; and

o. executing the Boolean query on the plurality of production documents in the production data.