System and method for automatic document classification in eDiscovery, compliance and legacy information clean-up

1. A computer implemented system for automatic document classification, the system comprising:

a processor;

a display device;

a memory in electronic communication with the processor; and

instructions stored in the memory, the instructions being executable by the processor to:

implement an extraction device to extract structural, syntactical, and semantic information from a document and normalize the extracted information,

wherein the extracted information includes named entities, properties of entities, noun-phrases, facts, events, sentiments, emotions, symbolism and concepts;

implement a machine learning device to generate a model representation for automatic document classification based on feature vectors built only from the normalized and extracted semantic information for machine learning and clustering; and

implement a classification device to select a non-classified document from a document collection, and via the extraction device extract the normalized structural, syntactical, and semantic information from the selected document, and generate via the machine learning device, the model representation of the selected document based on the feature vectors, and match the model representation of the selected document against the machine learning model representation to generate a document category, and classification for display on the display device to a user.

2. The system of claim 1 , wherein the extraction device employs text-mining, language identification, gazetteers, regular expressions, noun-phrase identification with part-of-speech taggers, and statistical models and rules, and is configured to identify patterns, and

the patterns include libraries, and algorithms shared among cases, and which can be tuned for a specific case, to generate case-specific semantic information.

3. The system of claim 1 , wherein the extracted information is normalized by using normalization rules, groupers, thesauri, taxonomies, and string-matching algorithms.

4. The system of claim 1 , wherein the model representation of the document is a term frequency-inverse document frequency (TF-IDF) document representation of the extracted information, and the clustering and machine learning includes a classifier based on support vector machines (SVM), rules-based classification, Non-Negative Matrix Factorization (NMF), Continuous Active Learning (CAL), Simple Passive Learning (SPL), and hierarchical clustering methods.

5. The system of claim 1 , wherein the document comprises document sections, including pages of a document, paragraphs of a document, chapters of a document, and text sections of a document.

6. The system of claim 5 , wherein the document sections are from a predetermined time-based period representing learned, and extracted temporal patterns.

7. The system of claim 1 , wherein the system employs precision and recall in combination with cross validation to measure a quality of classifiers or cluster algorithms based on a training set and documents classified that are not in the training set, and representing an increase or decrease of a quality of the system in time as well as indicative cost measurements for classification of additional documents manually or automatically.

8. The system of claim 1 , wherein the document comprises multi-media files.

9. A computer implemented method for automatic document classification, the method comprising:

providing a processor, a display device, a memory in electronic communication with the processor, and instructions stored in the memory, the instructions being executable by the processor to perform the steps of:

extracting with an extraction device structural, syntactical, and semantic information from a document and normalizing the extracted information,

wherein the extracted information includes named entities, properties of entities, noun-phrases, facts, events, sentiments, emotions, symbolism and concepts;

generating with a machine learning device a model representation for automatic document classification based on feature vectors built only from the normalized and extracted semantic information for machine learning and clustering; and

selecting with a classification device a non-classified document from a document collection, and via the extraction device extract the normalized structural, syntactical, and semantic information from the selected document, and generate via the machine learning device, the model representation of the selected document based on the feature vectors, and match the model representation of the selected document against the machine learning device representation to generate a document category, and classification for display on the display device to a user.

10. The method of claim 9 , wherein the extraction device employs text-mining, language identification, gazetteers, regular expressions, noun-phrase identification with part-of-speech taggers, and statistical models and rules, and is configured to identify patterns, and

the patterns include libraries, and algorithms shared among cases, and which can be tuned for a specific case, to generate case-specific semantic information.

11. The method of claim 9 , further comprising:

normalizing the extracted information using normalization rules, groupers, thesauri, taxonomies, and string-matching algorithms.

12. The method of claim 9 , wherein the model representation of the document is a term frequency-inverse document frequency (TF-IDF) document representation of the extracted information, and the clustering and machine learning includes a classifier based on support vector machines (SVM), rules-based classification, Non-Negative Matrix Factorization (NMF), Continuous Active Learning (CAL), Simple Passive Learning (SPL), and hierarchical clustering methods.

13. The method of claim 9 , wherein the document comprises document sections, including pages of a document, paragraphs of a document, chapters of a document, and text sections of a document.

14. The method of claim 13 , wherein the document sections are from a predetermined time-based period representing learned, and extracted temporal patterns.

15. The method of claim 9 , further comprising:

employing precision and recall in combination with cross validation to measure a quality of classifiers or cluster algorithms based on a training set and documents classified that are not in the training set, and representing an increase or decrease of a quality of the system in time as well as indicative cost measurements for classification of additional documents manually or automatically.

16. The method of claim 9 , wherein the document comprises multi-media files.

17. A computer program product for automatic document classification and including a processor, a display device, a memory in electronic communication with the processor, and instructions stored in the memory, the instructions being executable by the processor to perform the steps of:

extracting with an extraction device structural, syntactical, and semantic information from a document and normalizing the extracted information,

wherein the extracted information includes named entities, properties of entities, noun-phrases, facts, events, sentiments, emotions, symbolism and concepts;

generating with a machine learning device a model representation for automatic document classification based on feature vectors built only from the normalized and extracted semantic information for machine learning and clustering; and

selecting with a classification device a non-classified document from a document collection, and via the extraction device extract the normalized structural, syntactical, and semantic information from the selected document, and generate via the machine learning device, the model representation of the selected document based on the feature vectors, and match the model representation of the selected document against the machine learning device representation to generate a document category, and classification for display on the display device to a user.

18. The computer program product of claim 17 , wherein the extraction Currently amended employs text-mining, language identification, gazetteers, regular expressions, noun-phrase identification with part-of-speech taggers, and statistical models and rules, and is configured to identify patterns, and

the patterns include libraries, and algorithms shared among cases, and which can be tuned for a specific case, to generate case-specific semantic information.

19. The computer program product of claim 17 , further comprising:

normalizing the extracted information using normalization rules, groupers, thesauri, taxonomies, and string-matching algorithms.

20. The computer program product of claim 17 , wherein the model representation of the document is a term frequency-inverse document frequency (TF-IDF) document representation of the extracted information, and the clustering and machine learning includes a classifier based on support vector machines (SVM), rules-based classification, Non-Negative Matrix Factorization (NMF), Continuous Active Learning (CAL), Simple Passive Learning (SPL), and hierarchical clustering methods.

21. The computer program product of claim 17 , wherein the document comprises document sections, including pages of a document, paragraphs of a document, chapters of a document, and text sections of a document.

22. The computer program product of claim 21 , wherein the document sections are from a predetermined time-based period representing learned, and extracted temporal patterns.

23. The computer program product of claim 17 , further comprising:

employing precision and recall in combination with cross validation to measure a quality of classifiers or cluster algorithms based on a training set and documents classified that are not in the training set, and representing an increase or decrease of a quality of the system in time as well as indicative cost measurements for classification of additional documents manually or automatically.

24. The computer program product of claim 17 , wherein the document comprises multi-media files.