Document classification
View Patent ↗Provided are, among other things, systems, methods and techniques for classifying a collection of documents. A term is identified based on an indication of ability of the term's presence within a given document to predict whether the given document should be classified into an identified category. A document index is then queried using the identified term and, in response, search results that define a candidate set of documents are received. Finally, a classifier is applied to documents within the candidate set to determine which of the documents should be classified into the identified category.
1. A method comprising:
for a given category, receiving a positive set of training documents within the given category and a negative set of training documents not within the given category, by a processor of a computing device;
using a feature selector on the positive set and the negative set to determine a first set of features that are predictive for the given category, by the processor, each feature comprising a word or a phrase of words;
training a first classifier from the positive set and the negative set to assign a weight to each feature of the first set of features, by the processor;
after training the first classifier, querying a document index of a plurality of production documents for each feature of the first set of features to yield a sub-plurality of the production documents that are likely but not necessarily within the given category, by the processor, by formulating a query that includes the first set of features as weighted by the weights thereof to locate the sub-plurality of the production documents, where each of the sub-plurality of the production documents resulting from querying the document index has a total sum of the weights of the first set of features greater than a threshold; and
applying a second classifier that uses a second set of features greater in number than the first set to determine whether each production document of the sub-plurality is predicted to be within the given category, by the processor,
wherein the document index is queried to decrease a number of the production documents against which the second classifier is applied to just the sub-plurality of the production documents yielded by querying the document index,
and wherein the first classifier is one of a Naïve Bayesian classifier, a support vector machine classifier, and a logistic regression classifier.
2. The method of claim 1 , further comprising training the second classifier from the positive set and the negative set, by the processor.
3. The method of claim 1 , further comprising outputting the production documents of the sub-plurality that are predicted to be within the given category, by the processor.
4. The method of claim 1 , wherein the sub-plurality of the documents resulting from querying the document index includes the production documents that contain any of the first set of features.
5. The method of claim 1 , wherein some of the sub-plurality of production documents are not within the given category.
6. The method of claim 1 , wherein the sub-plurality of production documents is smaller in number than the plurality of production documents.
7. The method of claim 6 , wherein the sub-plurality of production documents that are within the given category is smaller in number than a total number of the sub-plurality of production documents.
8. A non-transitory machine-readable medium storing machine-executable process steps for classifying a collection of documents, said process steps comprising:
for a given category, receiving a positive set of training documents within the given category and a negative set of training documents not within the given category;
using a feature selector on the positive set and the negative set to determine a first set of features that are predictive for the given category, each feature comprising a word or a phrase of words;
training a first classifier from the positive set and the negative set to assign a weight to each feature of the first set of features;
after training the first classifier, querying a document index of a plurality of production documents for each feature of the first set of features to yield a sub-plurality of the production documents that are likely but not necessarily within the given category, by formulating a query that includes the first set of features as weighted by the weights thereof to locate the sub-plurality of the production documents, where each of the sub-plurality of the production documents resulting from querying the document index has a total sum of the weights of the first set of features greater than a threshold; and
applying a second classifier that uses a second set of features greater in number than the first set to determine whether each production document of the sub-plurality is predicted to be within the given category,
wherein the document index is queried to decrease a number of the production documents against which the second classifier is applied to just the sub-plurality of the production documents yielded by querying the document index,
and wherein the first classifier is one of a Naïve Bayesian classifier, a support vector machine classifier, and a logistic regression classifier.
9. The non-transitory machine-readable medium of claim 8 , wherein said process steps further comprise training the second classifier from the positive set and the negative set.
10. The non-transitory machine-readable medium of claim 8 , wherein said process steps further comprise outputting the production documents of the sub-plurality that are predicted to be within the given category.
11. The non-transitory machine-readable medium of claim 8 , wherein the sub-plurality of the documents resulting from querying the document index includes the production documents that contain any of the first set of features.
12. The non-transitory machine-readable medium of claim 8 , wherein some of the sub-plurality of production documents are not within the given category.
13. The non-transitory machine-readable medium of claim 8 , wherein the sub-plurality of production documents is smaller in number than the plurality of production documents.
14. The non-transitory machine-readable medium of claim 13 , wherein the sub-plurality of production documents that are within the given category is smaller in number than a total number of the sub-plurality of production documents.