Character strings in sample data are classified into groups of character strings with the same leading n characters (for example, "abc"). Then, one character string with the highest appearance frequency (the most frequently appearing character string) in the sample data is extracted from each group. The most frequently appearing character strings extracted from each group are registered in a dictionary as initial values in descending order of appearance frequency. Alternatively, character strings in sample data are classified into groups of character strings with the same hash value of leading n characters, the most frequently appearing character string is detected from each of the groups and the most frequently appearing character string is registered in the dictionary as an initial value.