Landscape search – beyond result lists

Gerald Landl is head of the IP-rights – Standards – Literature department at voestalpine Stahl GmbH and is running a patent agency - "Dipl. Ing. Gerald LANDL Patentagentur".

As a patent engineer with a degree in technical chemistry he was already  involved with the development of highly interlinked information systems during his university yearsexploring the needs and obstacles of company wide information systems. After some experience in R&D and being co-inventor for a high temperature protection coating at voestalpine Stahl GmbH, constantly involved in patent search, he implemented a corporate wide information system for standards and became head of the department IP-rights & Standards department in 2005. After implementing an IP-information system providing corporate wide service he and his colleguees focused the past three years on developing a semantic search tool using a graphic user interface (landscape search).

Abstract

In everyday industrial practice a permanent pressure to increase efficiency of methods and workflows and the reduction of skilled IR-resources lead to an ever demanding search for tools and aids. Traditionally, persons skilled in information retrieval would use a combination of patent classes, keywords, assignees, inventors and combinations thereof to reduce their search results to a bear- and workable number of documents which have to be analyzed in detail by IP-professionals. Commonly this is established by working through patent number lists, preferably with direct access to the full text documents. This way a high precision but not necessarily the best possible recall can be achieved. Especially in the case of new technical fields, which may represent a cross over of different classes the reduction to a proper number of hits is not always possible, a search for a freedom to operate will be extensively time consuming. Or in the case of a prior art search to find documents of an opposition or the like, the fully relevant documents can’t be found.

The search within very big numbers of documents to fulfill the above mentioned tasks can be accomplished quicker, with a higher precision and a recall which would find even the one relevant document needed for success. All this can be done using a graphical interface showing your results as a landscape, by showing similar documents neighbouring each other and the number of hits by increasing the height of the hills in a specific region. Even the development of a technical field can be monitored by comparing the landscapes based on the same search over time. In a test, using a restricted number of documents we were able to show that a set of documents which were extracted over hours by a skilled IR-specialist could be found within half an hour using the landscape method. 

The system was developed in cooperation between voestalpine Stahl GmbH and m2n in a R&D project the past three years. There are further tests still needed in the comparison with a traditional search and trainings in everyday work situations with the system need to be made to evaluate the reliability further and establish enough trust in workflows like this.