Document analysis system

A document analysis system comprising:

• a perplexity evaluation element configured to evaluate, for a plurality of documents, a perplexity of each document in accordance with a document generation probability in which the document is generated when topic counts for defining a topic model based on LDA as a document generation model are hypothetically specified as different values and each of word groups are specified by each of different random numbers;
• a first topic count determination element configured to accumulate frequencies at which the perplexity evaluated by the perplexity evaluation element first indicates a minimum value with respect to each of the word groups specified by each of the different random numbers and determine, as a first topic count, a particular topic count that indicates a highest cumulative frequency among the topic counts at which the perplexity indicates a minimal value;
• a second topic count determination element configured to accumulate frequencies at which the perplexity evaluated by the perplexity evaluation element indicates a smallest value with respect to each of the word groups specified by each of the different random numbers and determine, as a second topic count, a particular topic count that indicates a highest cumulative frequency among the topic counts at which the perplexity indicates a smallest value;
• and
• a reference topic count determination element configured to determine a reference topic count by combining the first topic count determined by the first topic count determination element and the second topic count determined by the second topic count determination element.