Systems and method for performing contextual classification using supervised and unsupervised training
Computerized systems and methods are disclosed for performing contextual classification of objects using supervised and unsupervised training. In accordance with one implementation, content reviewers may review training objects and submit supervised training data for preprocessing and analysis. The supervised training data may be preprocessed to identify key terms and phrases, such as by stemming, tokenization, or n-gram analysis, and form vectorized objects. The vectorized objects may be used to train one or more models for subsequent classification of objects. In certain implementations, preprocessing or training, among other steps, may be performed in parallel over multiple machines to improve efficiency. The disclosed systems and methods may be used in a wide variety of applications, such as article classification and content moderation.
1. A computer-implemented method for performing contextual classification of objects, the method comprising the following operations performed by one or more processors:
receiving supervised training data transmitted over a network from at least one content reviewer device;
identifying one or more training objects from the received supervised training data and further identifying a type of the one or more training objects and one or more key terms that define the one or more training objects, wherein identified types of the one or more training objects are at least one of: a training article, a training text, and a training image or video;
determining whether any key terms defining one or more of the identified training objects is abusive or non-abusive by comparing the one or more identified key terms and any determined abusive key terms and non-abusive key terms corresponding to the one or more training objects to detect matching key terms;
classifying the one or more training objects as abusive or non-abusive by assigning abusive or non-abusive labels to the one or more training objects determined to be defined by one or more detected matching key terms;
processing the supervised training data to form at least one vectorized object based on the classifying of the one or more training objects of the supervised training data;
training a plurality of models by applying a plurality of machine learning algorithms to each vectorized object of the supervised training data in parallel based on the processed supervised training data;
identifying an optimal model from among the trained plurality of models; and
classifying a candidate object of electronic content using the identified optimal model.
2. The computer-implemented method of claim 1 , wherein the at least one content reviewer is selected from the group consisting of an editor, an author, and a third-party user.
3. The computer-implemented method of claim 1 , wherein processing the supervised training data to form at least one vectorized object comprises performing at least one operation selected from the group consisting of stemming, tokenization, and n-gram analysis.
4. The computer-implemented method of claim 1 , wherein the plurality of machine learning algorithms comprises a support vector machine algorithm.
5. The computer-implemented method of claim 1 , wherein processing the supervised training data to form at least one vectorized object comprises processing the supervised training data using a plurality of machines operating in parallel.
6. The computer-implemented method of claim 1 , wherein:
receiving supervised training data transmitted over a network from at least one content reviewer comprises receiving at least one pre-defined tag selected by the at least one content reviewer as being associated with at least one training article; and
classifying a candidate object using the identified optimal model comprises automatically classifying a candidate article.
7. The computer-implemented method of claim 1 , wherein:
receiving supervised training data transmitted over a network from at least one content reviewer comprises receiving at least one training article and corresponding key terms selected by a content reviewer specifying whether at least one comment associated with an article is abusive; and
classifying a candidate object using the identified optimal model comprises automatically filtering at least one abusive comment.
8. A system for performing contextual classification of objects, comprising:
a memory configured to store supervised training data collected from at least one content reviewer device;
at least one processor configured to:
identify one or more training objects from the received supervised training data and further identifying a type of the one or more training objects and one or more key terms that define the one or more training objects, wherein identified types of the one or more training objects are at least one of: a training article, a training text, and a training image or video;
determine whether any key terms defining one or more of the identified training objects is abusive or non-abusive by comparing the one or more identified key terms and any determined abusive key terms and non-abusive key terms corresponding to the one or more training objects to detect matching key terms;
classify the one or more training objects as abusive or non-abusive by assigning abusive or non-abusive labels to the one or more training objects determined to be defined by one or more detected matching key terms;
process the supervised training data to form at least one vectorized object based on the classifying of the one or more training objects of the supervised training data;
train a plurality of models by applying a plurality of machine learning algorithms to each vectorized object of the supervised training data in parallel based on the processed supervised training data;
identify an optimal model from among the trained plurality of models; and
classify a candidate object of electronic content using the identified optimal model.
9. The system of claim 8 , wherein the at least one content reviewer is selected from the group consisting of an editor, an author, and a third-party user.
10. The system of claim 8 , wherein to process the supervised training data to form at least one vectorized object the at least processor is configured to perform at least one operation selected from the group consisting of stemming, tokenization, and n-gram analysis.
11. The system of claim 8 , wherein the plurality of machine learning algorithms comprises a support vector machine algorithm.
12. The system of claim 8 , wherein to process the supervised training data to form at least one vectorized object the at least processor is configured to process the supervised training data using a plurality of machines operating in parallel.
13. The system of claim 8 , wherein:
the memory is further configured to store at least one pre-defined tag selected by the at least one content reviewer in association with at least one training article; and
the at least one processor is further configured to classify a candidate object using the identified optimal model by automatically classifying a candidate article.
14. The system of claim 8 , wherein:
the memory is further configured to store at least one training article and corresponding key terms selected by a content reviewer specifying whether at least one comment associated with an article is abusive; and
the at least one processor is further configured to classify a candidate object using the identified optimal model by automatically filtering at least one abusive comment.
15. A computer-readable storage medium including instructions for performing contextual classification of objects, which, when executed by at least one processor, cause the processor to perform operations comprising:
collecting supervised training data from at least one content reviewer device;
identifying one or more training objects from the received supervised training data and further identifying a type of the one or more training objects and one or more key terms that define the one or more training objects, wherein identified types of the one or more training objects are at least one of: a training article, a training text, and a training image or video;
determining whether any key terms defining one or more of the identified training objects is abusive or non-abusive by comparing the one or more identified key terms and any determined abusive key terms and non-abusive key terms corresponding to the one or more training objects to detect matching key terms;
classifying the one or more training objects as abusive or non-abusive by assigning abusive or non-abusive labels to the one or more training objects determined to be defined by one or more detected matching key terms;
processing the supervised training data to form at least one vectorized object based on the classifying of the one or more training objects of the supervised training data;
training a plurality of models by applying a plurality of machine learning algorithms to each vectorized object of the supervised training data in parallel based on the processed supervised training data;
identifying an optimal model from among the trained plurality of models; and
classifying a candidate object of electronic content using the identified optimal model.
16. The computer-readable storage medium of claim 15 , wherein the at least one content reviewer is selected from the group consisting of an editor, an author, and a third-party user.
17. The computer-readable storage medium of claim 15 , wherein the operations performed by the at least one processor further comprise performing at least one operation selected from the group consisting of stemming, tokenization, and n-gram analysis.
18. The computer-readable storage medium of claim 15 , wherein the plurality of machine learning algorithms comprises a support vector machine algorithm.
19. The computer-readable storage medium of claim 15 , wherein the operations performed by the at least one processor further comprise processing the supervised training data using a plurality of machines operating in parallel.
20. The computer-readable storage medium of claim 15 , wherein the operations performed by the at least one processor further comprise:
collecting at least one pre-defined tag selected by the at least one content reviewer as being associated with at least one training article; and
automatically classifying a candidate article.
21. The computer-readable storage medium of claim 15 , wherein the operations performed by the at least one processor further comprise:
collecting at least one training article and corresponding key terms selected by a content reviewer specifying whether at least one comment associated with an article is abusive; and
automatically filtering at least one abusive comment.