Land register entries are the backbone of all real estate and construction planning. They contain sensitive information about ownership, easements and mortgages. The manual recording and processing of this data is not only time-consuming, but also harbors a high potential for error. With AI-based automated data extraction, however, this information can be extracted precisely and efficiently from the land register entry.
This article describes in a concrete use case the great added value that lies in the immediate availability of structured information that can be integrated directly into planning processes. This speeds up decision-making and significantly reduces the risk of planning errors.
Initial situation: Check land register extracts
Renewable energies are securing a prominent position in political debates and social perspectives for the future. The planning and construction of wind turbines is an important part of these discussions and represents a complex task that is accompanied by numerous legal requirements. Land register entries, building encumbrances and building rights play a central role here. Efficient management of these documents is necessary to ensure the success of such projects. This is where the importance of an automated document processing solution such as the Helm & Nagel GmbH, with the AI software Konfuzio.
Challenge: Automatically recognize information in land register excerpts
The challenge in planning wind turbines lies in the management and processing of extensive and complex legal documents. Land register entries containing important information such as owner data, easements and mortgages need to be carefully analyzed. The manual process by which employees in this use case transfer the information into Excel spreadsheets in order to then integrate it into a geographic information system (GIS) is time-consuming and prone to errors. An automated solution is therefore required to make these processes more efficient and minimize the source of errors.
Goal: Automatic processing of land register extracts
The objective is to implement a solution that automates the processing and management of documents relating to land registries, project areas, building charges and building rights. This solution should enable relevant information to be extracted quickly and accurately and put into a structured form so that it can be processed directly in internal processes.
A key component of this project is the Proof of Concept (POC)which shows how the automatic reading of the land register entry can be implemented. The aim of this POC is to extract specific fields from the land register entries so that they can be used efficiently in further project planning. The fields to be extracted include
- General information on the land register:
- Land register of
- Date of moving out
- Leaf
- Local court
- Inventory:
- Serial number of the parcel
- Former number of the parcel
- District
- Hallway
- Parcel
- Economic type
- First department:
- Sequence number of the entry
- Owner
- Corresponding serial numbers of the parcels (partly indicated individually, partly with 1-5 or 1,2,3)
- Basis of registration
- Second department:
- Sequence number of the entry
- Serial number of the parcel
- Loads & restrictions
- Third division:
- Sequence number of the entry
- Serial number of the parcel
- Amount
- Mortgage, land charge, annuity debt
This detailed extraction of land register entry data makes it possible to integrate the relevant legal information directly into the planning processes.
Analysis and preparation
A well thought-out project structure is required for the automated readout of the land register entry. The relevant fields from the land register entries in Konfuzio are stored in so-called Labels and LabelSets organized. This categorization enables systematic processing of the documents:
- Labels are specific categories or tags that are assigned to individual data points in a document to identify certain information, e.g. "owner" or "parcel number".
- LabelSets are groups of labels that belong together thematically or functionally, e.g. all labels that contain information on the "inventory" in the land register.
This structuring helps the AI to precisely identify and extract the relevant data.
In the next step, the documents are annotated, i.e. marked, using these labels. These Annotations serve as training data for the AI. The data set is continuously expanded through repeated training and the addition of new data, which improves the AI's extraction performance.
After completing the training, the trained Extraction AI automatically extract the defined fields from the land register entry and provide the information in a structured manner. The land register entry is uploaded in the form of a PDF file, whereupon the AI automatically extracts the data.
After extraction, the Konfuzio platform takes over the annotations of the data and transfers them into a JSON format using a web-based interface (REST API). This enables seamless integration of the data into downstream processes, such as transfer to a geographic information system (GIS) or other data processing systems.
Annotation refers to the marking and labeling of specific data or information in a document. In AI-supported data processing, these markers are used to teach the AI which data is relevant. In this way, the AI automatically recognizes similar information in other documents and can also extract it.
Conversion: Land register extract KI
An iterative approach and adjustments that are specifically tailored to the requirements of the land register entry significantly increase the extraction AI and the quality of the data set. The implementation takes place in the following steps:
- Definition of the project structure
Define a clear structure for the project that identifies all relevant fields in the land register entries and systematically organizes them into labels and label sets. This structure enables efficient and orderly processing of the documents. - Data preparation
Upload the land register entries to the system and prepare them for training, ensuring precise and consistent annotation of the data. This care ensures high data quality, which is crucial for AI training. - Model training
The model training takes place with the carefully annotated data, whereby several training runs are carried out. This allows the AI to be optimally calibrated to the specific requirements of the land register entries. - Model fine-tuning
After the initial training, the model is refined. By adjusting the Hyperparameters and optimization based on the initial results, the extraction accuracy is further increased. - Validation
The trained model is then validated with test data. This validation shows how well the AI can read the land register entry in practice. - Optimization
Based on the validation results, final adjustments and optimizations are made to prepare the extraction AI for productive use and maximize the readout accuracy.
Land register extract OCR and AI performance
The results are based on a dataset distribution of around 70 % training data and 30 % test data. In the first step, the AI was trained to create a basis for the automated creation of annotations. These were then checked by a Konfuzio expert and improved where necessary. The result was a significantly faster annotation process and a high-quality data set. Another important factor for the performance of the AI is the quality of the training data. Consistent data quality is crucial to ensure the quality of future annotations.
Conclusion
The implementation of an automated solution for reading data from land register entries proves to be a decisive step towards increasing efficiency in the planning of wind turbines in this use case. By automating the capture and management of legally relevant documents, internal processes can be significantly accelerated and the accuracy of data transmission improved. The combination of careful analysis, robust artificial intelligence for data extraction and the possibility of integration via an interface creates a powerful, flexible and future-oriented solution.
This advanced solution from Konfuzio is instrumental in paving the way for the successful realization of wind energy or related projects and meeting the demands of an increasingly complex legal landscape.
Take Contact and explore the automation potential for your specific application.
FAQ Land register entry
A land register entry is the official entry of a property or plot of land in the land register by the land registry, often in conjunction with a land charge in the case of construction financing.
A land register entry is required by buyers and sellers of real estate or land in order to have the change of ownership and land charge officially entered in the land register.
The costs for a land register entry, including notary and land register extract, vary, but often amount to several hundred euros - depending on the value of the property or land and the amount of the land charge.
An entry in the land register can take several weeks, depending on the processing time at the notary and land registry as well as the department responsible for the entry in the land register.
EN 
DE 
ES 
FR