Topic predictions based on natural language processing of large corpora

A tangible, non-transitory, machine-readable medium storing instructions that when executed by one or more processors effectuate operations comprising:

• obtaining, with one or more processors, a corpus having a plurality of time-stamped documents having unstructured natural language text;
• detecting, with one or more processors, topics in the unstructured natural language text and, in response, associating with each of the documents one or more topic scores by which the topics are detected, wherein:
• detecting topics comprises determining the topic scores;
• the topic scores are indicative of whether an associated document in the corpus pertains to a topic corresponding to the respective topic score;
• detecting topics comprises detecting whether the documents pertain to any of a plurality of topics such that a determination is made for each permutation of topic-and-document pair;
• determining, with one or more processors, temporal-metrics of the topics based on time-stamps of documents to which the topics pertain, wherein:
• determining temporal-metrics comprises aggregating topic scores over time according to the time-stamps of the documents to form aggregated topic scores;
• determining temporal-metrics comprises determining a measure of a peak aggregated topic score for at least some of the topics;
• and
• determining temporal-metrics comprises determining a measure of persistence for at least some of the topics;
• predicting, with one or more processors, based on the temporal-metrics, whether a given topic among the plurality of topics will appear in future time-stamped documents having unstructured natural language text;
• and
• storing, with one or more processors, a result of the prediction in memory.