PaperlayerView exact revision on arXiv

Source-linked AI summary

A natural-language-based approach to intelligent data retrieval and representation for cloud BIM

Jia-Rui Lin, Zhen-Zhong Hu, Jian-Ping Zhang, Fang-Qiang Yu

arXiv:2411.09951v1cs.HC

TL;DR

Large, multidisciplinary BIM datasets make it difficult to retrieve focused information, particularly through limited mobile interaction. The paper combines cloud storage and query acceleration with NLP-based keyword and constraint extraction, IFD-to-IFC mapping, graph-based retrieval, and structured result representation. A Kunming Airport prototype demonstrates the approach, while the authors identify limitations in sentence complexity, concept coverage, representation, and mobile display space.

  • Problem

    Growing BIM data volumes make it difficult for users to obtain focused information from integrated AEC project models.

  • Method

    The approach combines cloud BIM storage, NLP-based keyword and constraint extraction, IFD mapping to IFC entities or attributes, graph-based retrieval, and structured result representation.

  • Results

    The Kunming Airport prototype correctly processed natural-language intentions, retrieved and represented data, and provided multi-aspect feedback through animations, charts, list views, and a timeline dashboard.

  • Takeaways & Limitations

    The approach can simplify information requests and enhance the value of BIM by supporting natural-language retrieval and multiple representations.

  • Takeaways & Limitations

    The approach currently supports only simple sentences, while keyword extraction, concept mapping, mobile representation, and custom configuration remain constrained.

Abstract

from arXiv · show

As the information from diverse disciplines continues to integrate during the whole life cycle of an Architecture, Engineering, and Construction (AEC) project, the BIM (Building Information Model/Modeling) becomes increasingly large. This condition will cause users difficulty in acquiring the information they truly desire on a mobile device with limited space for interaction. To improve the value of the big data of BIM, an approach to intelligent data retrieval and representation for cloud BIM applications based on natural language processing was proposed. First, strategies for data storage and query acceleration based on the popular cloud-based database were explored to handle the large amount of BIM data. Then, the concepts keyword and constraint were proposed to capture the key objects and their specifications in a natural-language-based sentence that expresses the requirements of the user. Keywords and constraints can be mapped to IFC entities or properties through the International Framework for Dictionaries (IFD). The relationship between the user's requirement and the IFC-based data model was established by path finding in a graph generated from the IFC schema, enabling data retrieval and analysis. Finally, the analyzed and summarized results of BIM data were represented based on the structure of the retrieved data. A prototype application was developed to validate the proposed approach on the data collected during the construction of the terminal of Kunming Airport, the largest single building in China. With this approach, users can significantly benefit from requesting for information and the value of BIM will be enhanced.

1 INTRODUCTION

BIM adoption and interdisciplinary integration are increasing the volume of AEC project information, making focused retrieval difficult. The paper proposes cloud-based, NLP-driven retrieval and representation to process natural-language requests and present relevant BIM results.

  • BIM integrates information and stakeholders across the AEC project life cycle, while adoption among AEC participants and companies continues to increase.
  • IFC-based methods support interoperability through a comprehensive building-data representation and defined rules and protocols.
  • Growing BIM data volumes make it difficult for users, including experts, to obtain concentrated and useful information without tedious manual manipulation.
  • The proposed framework stores and manipulates BIM data in the cloud, extracts natural-language keywords, maps them to IFC entities or attributes, and represents retrieved results as charts, tables, animations, or combinations.
  • The paper organizes its validation around a Kunming Airport case study after presenting cloud storage, NLP-based keyword mapping, and information retrieval and representation processes.

1. Interoperability based on IFC and IFD

IFC provides a widely used schema and exchange protocol for interoperable BIM information. IFD complements IFC by supplying controlled, multilingual terminology that maps concepts to IFC entities and attributes.

  • IFC represents building data and defines rules and protocols intended to address inadequate interoperability between software systems.
  • IFC-based systems and servers have supported design-information management, information sharing, extraction, and integration.
  • IFD supplies a controlled, multilingual construction vocabulary and maps concepts to IFC entities and attributes.

2. Cloud and BIM

As BIM grows, stand-alone information manipulation becomes more difficult, while cloud computing offers a basis for handling and sharing large BIM datasets. Existing cloud BIM products emphasize sharing, collaboration, and visualization but provide limited detail about storage and access extensibility.

  • Larger BIM models make information use and sharing through stand-alone systems more difficult, increasing the relevance of cloud computing for information manipulation.
  • Cloud BIM can enhance multidisciplinary collaboration, web-based access to information and resources, and handling of large datasets through an extensible platform.
  • Existing cloud BIM products mainly focus on data sharing, collaboration, and visualization, while their storage and access mechanisms are difficult to extend.

3. Data retrieval and visualization

Prior BIM retrieval work includes domain-specific query systems and graph-based approaches, while visualization research provides multiple ways to express information. These foundations motivate intelligent natural-language retrieval and structured representation for diverse BIM stakeholders.

  • BIM retrieval approaches include domain-specific interfaces such as BIMQL and graph-based methods for accessing BIM information.
  • Visualization studies address categorical and time-oriented data, as well as multiple design representation subsystems and their interactions.
  • IFC's object-oriented semantic schema provides a foundation for BIM interoperability, retrieval, and analysis as accumulated data continues to grow.
  • NLP offers established computerized methods and tools for analyzing text in different languages.
  • The paper identifies demand for intelligent retrieval and representation that supports non-experts and increases the value of cloud BIM data.

3 CLOUD-BASED DATA STORAGE

The cloud-based storage framework uses MongoDB and MapReduce to store, serialize, and query IFC-based BIM data at scale. Serialization strategies separate embedded, referenced, and geometric data, while pre-joining related collections reduces query overhead.

  • Cloud-based database: Cloud BIM data are stored in MongoDB clusters using IFC-format documents, with MapReduce supporting distributed analysis and manipulation.The framework uses cluster metadata, NoSQL databases, and MapReduce functions to support fast access to information.
  • Cloud-based database: MongoDB was selected for scalability, MapReduce support, file storage, indexing, and flexible querying.Its sharding and replication distribute and synchronize data across servers, while GridFS stores large files with metadata.
  • Serialization strategies: MongoDB stores same-type objects as BSON documents, embedding simple attributes and representing complex attributes as nested sub-documents or references.Most entity properties are embedded for flexible and fast access, while some relational objects are stored in separate collections.
  • Serialization strategies: IFC entities are serialized with different strategies: core objects and relationships become documents, resource-layer data are embedded, and geometric data use separate collections or cached display formats.The design avoids self-reference, deep nesting, and inefficient retrieval by choosing embedding, identifiers, or separate storage according to entity type.
  • Query acceleration: MapReduce-based pre-joining replaces two client-side queries with one query for related entities stored in different collections.The approach creates temporary entities through map functions and merges matching data through reduction before final storage.
  • Query acceleration: 55% to 67% of the original query time was required for collections containing 1,000,000 and 10,000 documents with pre-joining.The reported tests indicate lower query time than the original approach, providing a basis for efficient BIM data retrieval.

4 KEYWORD EXTRACTION AND MAPPING

The approach parses natural-language BIM requests into keywords and constraints, then maps those concepts to IFC entities or attributes through IFD. Syntactic structure and domain terminology guide extraction, disambiguation, and mapping.

  • 4.1 Main process: NLP processing uses tokenization, tagging, parsing, classification, and mapping to convert user input into IFC-linked keywords.Tokenization segments words; tagging labels parts of speech; parsing identifies relationships; classification separates words mapped to entities or attributes; mapping uses IFD–IFC relationships.
  • 4.2 Keyword extraction based on NLP: Keywords represent user-focused objects, while adjective and preposition phrases provide constraints and their ordering specifies the requested relationship.For “quantity of beams of second and third storey,” the extracted concepts are “quantity,” “beam,” and “storey,” with storey values constraining beams.
  • 4.2 Keyword extraction based on NLP: When adjacent noun-tagged leaves share a parent, dependency analysis identifies the keyword and its constraining term.In “construction progress of the check-in zone,” dependency analysis selects “progress” as the keyword constrained by “construction.”
  • 4.3 IFD-based keyword mapping: The IFD Library defines semantically described concepts and properties with globally unique identifiers that establish links to IFC entities.IFD concepts include subjects and characteristics, while relations such as “assigns properties” connect them; IFC model relationships are established through concept GUIDs.
  • 4.3 IFD-based keyword mapping: Keyword mapping is extended through synonym and multilingual terminology support so linguistic variants can resolve to the same construction concept.The prototype relates terms such as “girder” and “beam” to the same “beam” concept.

5 DATA RETRIEVAL AND REPRESENTATION 5.1 Relation finding and constraint transforming for data retrieval

The retrieval process searches an IFC-schema graph to connect requested entities and incorporates property and logical constraints. Retrieved BIM data are then classified, summarized, and represented using formats matched to their structure.

  • 5.1 Relation finding and constraint transforming for data retrieval: An IFC schema converted into a graph enables path search between entities, attributes, and inheritance relations.Entities and defined types become nodes, while attributes and inheritances become edges; algorithms such as Dijkstra can find paths connecting requested IFC entities.
  • 5.1 Relation finding and constraint transforming for data retrieval: Found paths encode entity relations, while value constraints and logical relations are added to retrieve the correct database records.For example, IfcRelContainedInSpatialStructure expresses that beams belong to a building storey, while property values further restrict the result.
  • 5.1 Relation finding and constraint transforming for data retrieval: Logical “and” and “or” expressions become set operations or database predicates depending on whether they connect keywords or adjunctive constraints.“Quantity of beams and columns” uses union of retrieved element sets, whereas “second or third storey” becomes an OR predicate on storey names.
  • 5.2 Data analysis and representation: Retrieved data can be classified by attribute values and summarized by group to calculate quantities or element totals.Beam records may be grouped by ObjectType, and summaries can calculate the quantity or total number of elements in each group.
  • 5.2 Data analysis and representation: Representation formats are selected according to data structure, including plain text, charts and tables, grouped charts, net graphs, and 3D visualization.Single values use plain text; arrays use charts and tables; trees use grouped charts; net structures use net graphs; geometry-related data use 3D views.
  • 5.2 Data analysis and representation: Time-dependent data are shown as 4D animations, while geometry linked to data is colored by the associated values and unrelated geometry is hidden.The representation rules support sequence visualization and selective display of BIM geometry.

6 CASE STUDY OF THE TERMINAL OF KUNMING CHANGSHUI AIRPORT

The Intelli-BIM prototype integrated design and construction BIM data in a cloud database and retrieved, analyzed, and represented user-requested information from natural-language queries. The Kunming Airport case showed multi-aspect feedback through charts, lists, animations, and a timeline dashboard, while query performance and sentence complexity remained bounded by stated implementation limitations.

  • Case study setup: The Intelli-BIM prototype was tested on BIM data collected during construction of the Kunming Airport terminal.The terminal model integrated approximately 1,100,000 objects and about 50 GB of BIM, schedule, quantity, picture, and document data.
  • Case study setup: Design, construction, quantity-takeoff, and schedule information were integrated and stored in an IFC-oriented MongoDB cloud database.The construction BIM combined exported IFC files, planned schedules, construction simulation information, and quantity data before cloud storage.
  • Query demonstration: The query “construction progress of the check-in zone” was processed through NLP and IFD mapping to retrieve and represent related data from the cloud database.The retrieved information was presented in multiple coordinated forms based on the proposed approach.
  • Query demonstration: Feedback combined a resource diagram, construction-schedule list, construction-process animation, and timeline dashboard for important tasks.The representations were matched to the types of information returned by the query.
  • Query demonstration: Charts used colored bars, line strips, and double Y-axes to represent resource types and quantities across short and long usage periods.The configured logarithmic and common axes supported data distributed across a large range.
  • Query demonstration: Collapsible task lists and timeline dashboards enabled overall views of construction tasks, milestones, snapshots, durations, contractors, and resources.The study reports that combining animations, charts, list views, and dashboards provided abundant feedback on the requested information.
  • Performance and limitations: Query time ranged from 1.5s to 3.5s, with keyword extraction and mapping taking about 0.5s.The investigation attributes remaining variation to data retrieval and representation, keyword count, NLP tools, IFD coverage, and IFC-schema graph relationship finding.

7 CONCLUSION AND FUTURE WORKS

The study combines cloud BIM storage, natural-language processing, IFD-based mapping, IFC-schema graph traversal, and format-aware representation to retrieve information for non-expert users. The authors report successful implementation in the case study but identify limited sentence complexity, representation configurability, IFD coverage, mobile space, and processing speed as future-work boundaries.

  • Conclusion: The proposed framework uses MongoDB cloud storage, MapReduce pre-joining, NLP keyword and constraint extraction, IFD mapping, IFC-schema graph relations, and format-aware result representation.Results are classified and summarized as tables, charts, animations, or combinations of these formats.
  • Conclusion: The approach is intended to facilitate natural-language information requests for non-experts and enhance the value of BIM.The authors state that it may also be adopted in facility management and cost management applications.
  • Future works: The current representation is not feasible for custom configuration, and only simple sentences are supported.Complex sentences containing verbs, operators, or complex calculations are not yet supported.
  • Future works: The approach remains constrained by IFD concept coverage, keyword-extraction and mapping bottlenecks, limited mobile display space, and absent full-text search for unstructured documents.Suggested improvements include enriching IFD, parallelizing NLP, caching frequent queries, improving mobile representations, and integrating document search.
Loading 2411.09951v1…