Intelligent patent analysis through the use of a neural network: experiment of multi-viewpoint analysis with the MultiSOM model

Colloque avec actes et comité de lecture. internationale.International audienceThe main area of this paper concerns the neural methods for mapping scientific and technical information (articles, patents) and for assisting a user in carrying out the complex process of analysing large quantities of such information. In the procedure of information analysis, like in the domain of patent analysis, the complexity of the studied topics and the accuracy of the question to be answered may often lead the analyst to partition his reasoning into viewpoints. Most of the classical information analysis tools can only manage an analysis of the studied domain in a global way. The information analysis tool that will be considered in our study is the MultiSOM tool whose core model represents a significant extension of the classical Kohonen SOM neural model. The MultiSOM neural-based tool introduces the concepts of viewpoints and dynamics into the information analysis with its multi-maps displays and its inter-map communication process. The dynamic information exchange between maps can be exploited by an analyst in order to perform cooperative deduction between several different analyzes that have been performed on the same data. The paper demonstrates the efficiency of a viewpoint-oriented-analysis as compared to a global analysis in the domain of patents. Both objective and subjective quality criteria are taken into account for quality evaluation. The experimental context of the paper is constituted by a patent database of 1000 patents related to oil engineering. The patents structure and the patents field semantics are firstly exploited in order to generate different viewpoints corresponding to different areas of interest for the analysts. In the experiment the selected viewpoints correspond to uses, advantages, patentees, and titles subfields of the patents. The indexing vocabulary of each viewpoint is automatically extracted of its related textual contents in the patents through a full text analysis. The resulting vocabulary is then used to rebuild patents descriptions regarding each viewpoint. These descriptions are finally classified through the unsupervised MultiSOM algorithm resulting in as much different maps as viewpoints. A fifth "global viewpoint" which represent the combination of all the specific ones is also considered in order to perform our comparison between a global classification mechanism and a pure viewpoint-oriented classification mechanism