BIM search engine: Exploiting interrelations between objects when assessing relevance
2018-04-05T09:58:17Z (GMT) by
An increasing amount of information is packed into Building Information Models (BIMs), with the 3D geometry intended to serve as a central index leading to other information. The Three-Dimensional Information Retrieval (3DIR) project investigated information retrieval from such environments, with the aim of developing a search engine for searching and retrieving information from a building model. Here, the 3D model of the building can be exploited to formulate queries, compute the relevance of information items to a given query, and visualize search results. The focus of this paper is the computing of relevance. Literature in BIM/CAD and information retrieval was reviewed as a precursor to developing the search engine. Based on earlier research which identified the needs and aspirations of the users of BIMs, a graph theoretic formulation is proposed here to inform the emerging retrieval mechanisms of a BIM search engine. This formulation distinguishes between 3D and textual information in the model (the vertices in the graph), and between different types of relationships linking model objects (the edges in the graph). The value is tested of exploiting a 3D object’s relations to other 3D objects when assessing that object’s relevance to a query. For example, if a user is searching for “glazing door internal wall”, such a holistic/contextual search would rate the relevance of a “glazing panel” object more highly if it was touching “internal wall” or “door” objects. This notion was tested using an Autodesk Revit model from an architectural industry partner, augmented with the 3DIR search toolset. The model contained just under 7k 3D elements. Relationships between the objects were either hosting, touching or intersecting relationships. A comparison of the retrieval performance for a handful of test queries with and without this holistic/contextual search function does not decisively highlight the benefit but demonstrates the promise of this approach particularly for more complex multiple search term queries, as well as the value of the underlying graph theoretic formulation for studying and developing such systems.