A method for forecasting a rate of improvement of discrete technology domains by combining a predictive model and technology model, comprising:
- creating a technology model that reflects a plurality of technology domains, by selecting a first plurality of patent classes from a first patent class type and a second plurality of patent classes from a second patent class type to form a plurality of class pairs, each of which reflects a potential technology domain;
- for each class pair in the plurality of class pairs:
- querying a database of patents to identify patents lying in both the selected first patent class and the selected second patent class;
- determining a threshold overlap standard based on an expected number of patents lying in both the selected first patent class and the selected second patent class by calculating a joint probability of any single patent appearing in both the selected first patent class and the selected second patent class if the pair was selected at random and was technologically unrelated and multiplying said probability by the total number of patents in the patent system to obtain an expected number of patents in any randomly selected unrelated class pair;
- comparing the number of said identified patents in the class pair with the threshold overlap standard, and
- if the number of said identified patents in the class pair is below the threshold overlap standard, determining that the class pair is unlikely to encompass a meaningful technology domain and discarding the class pair,
- if the number of said identified patents in the class pair is equal to or below the threshold overlap standard, determining that the class pair is likely to encompass a meaningful technology domain and retaining the class pair;
- updating the technology model, by calculating and storing, for each class pair, an average patent centrality value for the said identified patents within each class pair by analyzing a graph representation of said identified patents with a normalized search path node pair index and training a prediction model for the rate of improvement of technology domains;
- and
- applying the prediction model to obtain an estimated improvement rate based on said calculated average centrality value for each of the class pairs.