Construction of text classifiers

A system comprising:

• one or more computers configured to perform operations comprising:
• receiving a collection of documents;
• for each document of the collection of documents:
• breaking the document into pieces of text, and
• extracting n+k-grams from each piece of text separately such that each n+k-gram does not overlap;
• using the extracted n+k-grams for each document as a collection of candidate phrases for a given topic;
• assigning weights to the candidate phrases;
• and
• generating a linear classifier using the weighted candidate phrases, wherein the linear classifier varies the weights for each phrase candidate depending on the length of a document being classified.

A computer program product, stored on a non-transitory computer readable medium, comprising instructions that when executed on a server cause the server to perform operations comprising:

• receiving a collection of documents;
• for each document of the collection of documents:
• breaking the document into pieces of text, and
• extracting n+k-grams from each piece of text separately such that each n+k-gram does not overlap;
• using the extracted n+k-grams for each document as a collection of candidate phrases for a given topic;
• assigning weights to the candidate phrases;
• and
• generating a linear classifier using the weighted candidate phrases, wherein the linear classifier varies the weights for each phrase candidate depending on the length of a document being classified.