Results (
Thai) 1:
[Copy]Copied!
4. Discussion of results: comparison of clustering analysis and classification-based analysisAs mentioned above, the PackMOLETM prototype is able to export data regarding the outcome of a cluster- ing session to an external software program for further processing, namely a spreadsheet such as, for instance, Microsoftâ Excel.Using this feature, the first 10 most numerous Der- went Classes [11] were extracted from both time slices, as shown in Table 9, to study the patenting activities of our target company with a classical grouping technique based upon patent classifications, and then to compare the similarities and differences of this standard analysis with the patent maps previously obtained through the clustering sessions.Clearly, even though the two different analyses more or less show the same overall trends, the patent maps obtained through text mining are easier to understand, in part because they are presented in graphical rather than textual or tabular form.In fact, the patent maps generated by text clustering allow for a better overview of the relationships between the different areas of patent activity, at the same time avoiding the work involved in using different, more de- tailed patent classifications, such as for example the IPC. In particular, the IPC was felt to be either too broad (at the class/subclass level) or too detailed (at the group/ subgroup level) to effectively carry out an optimal patent portfolio analysis.Regarding the Derwent Classification, it is to be noted that the majority of retrieved Derwent Classes belonged to the Engineering sections P and Q where each Derwent Class automatically corresponds to an exact, predetermined range of IPCs [11].This subdivision scheme did not always prove effec- tive: for instance, in some extreme cases a few patents were classified with different IPCs, even if they clearly referred to inventions sharing the same subject matter, 5 and their respective IPCs were spaced so far apart from each other that they were assigned different Derwent Classes as well. On the contrary, the PackMOLETM prototype was able to correctly group these patents into the same clusters.In any event, the good performances exhibited by the PackMOLETM prototype in correctly grouping patent documents were probably greatly enhanced by the high quality of Derwent abstracting.Another strong point shown by the PackMOLETM prototype was to provide the analyst with the ability to correctly identify, by comparing the two patent maps shown in Figs. 3 and 4, the changes in patent activities in the different business areas of our target company, as well as the subtle dynamics related to technological developments and spin-offs, that were not otherwise immediately detectable through the classical analysis: Indeed, by only relying on the results shown in Table 9, one might have deduced quite opposite conclusions.Finally, it is worth noting that the validated patent maps shown in Figs. 3 and 4 do not exactly represent the complete patent portfolio of our target company, as a few clusters were deemed to be invalid and thus re- moved, as previously mentioned.Rather, the information contained in the validated patent maps was thought to constitute a fairly good picture of the patenting trends of our target company.
Being translated, please wait..
