NATURAL LANGUAGE PROCESSING KEYWORD ANALYSIS
As disclosed herein, a method for generating a natural language processing query includes receiving one or more documents, wherein each document comprises a set of words, processing the one or more documents and the sets of words to provide a document content matrix V, a word feature matrix W, and a document feature matrix H, forecasting values for each entry of the word feature matrix and the document feature matrix over a selected time interval and a selected set of domains to provide a forecasted word feature matrix W′ and a forecasted document feature matrix H′, calculating a set of coefficients for forecasted document feature matrix H′ such that V=W′*H′, determining a rank for each word of the sets of words according to the calculated set of coefficients, and generating one or more queries according to the determined ranks for each word of the set of words.
1 . A computer implemented method for generating natural language processing queries, the method comprising:
receiving one or more documents, wherein each document comprises a set of words;
receiving a set of feature information, wherein the set of feature information indicates one or more features, and wherein the indicated features are the features that are analyzed within each of the documents;
processing the one or more documents and the sets of words to provide a document content matrix V corresponding to the one or more documents and the sets of words, a word feature matrix W corresponding to a set of selected features and the sets of words, and a document feature matrix H corresponding to the one or more documents and the set of selected features;
using stochastic gradient descent to calculate a feature value for each feature for the sets of words, wherein the feature value is the appropriate word feature matrix entry;
forecasting values for each entry of the word feature matrix and the document feature matrix over a selected time interval and a selected set of domains to provide a forecasted word feature matrix W′ and a forecasted document feature matrix H′ by calculating predicted values for each entry according to trends displayed over time for each word feature and trends displayed in the selected set of domains;
calculating a set of coefficients for forecasted document feature matrix H′ such that V=W′*H′;
determining a rank for each word of the sets of words according to the calculated set of coefficients by calculating an average of the coefficients for each word, and sorting the words by average; and
using a query template to generate one or more queries according to the determined ranks for each word of the set of words and according to one or more word types.