The purpose of the plugin is to recommend already existing categories/tags on newly created OSM entities. When a user creates a new object on the map, the plugin provides a list of the most relevant existing categories (expressed as tags). The tags can be added automatically to the selected OSM entity.
The classification process is carried out using Support Vector Machines by analyzing spatial entities into training features. The recommended tags are provided by the produced SVM model from the training process See Below.
Training an SVM model
For training recommendation models, after clicking the “Train a Model” button, the training configuration for the SVM model pops up. Here input:
- OSM file on which the model will be trained (default is the opened file in JOSM).
- SVM parameters:
- If C is not provided, the system runs a cross-validation process against a set of different c parameters and evaluates the best C.
- Maximum thresholds for the textual training features (top-k or max-Frequency).
- You can train a model by the editing history of a specific user, by area or by history, providing the additional info.
- Click Accept and Train Model button to start the training process.
Download ready to use SVM models
In the following link you can download ready to use SVM models for several cities in the world: SVM Models
After you download the compressed file for your preferred city, simply extract the "OSMRec_models" folder in the same folder of the OSM file you are going to edit with JOSM. Now you are ready to get recommendations from OSMRec without having to train a model!
For the category recommendation process on newly inserted spatial entities, click the "Add Recommendation" button from the initial toggle box.
The system loads the appropriate recommendation SVM model and a top-10 list of recommended categories appear.
OSMRec allows to choose a custom model or combine different training models.
By clicking the “Model Settings” button you can choose to use a single model or combine several SVM models with configurable weights.
Note: When opening the plugin, it creates a directory "OSMRec_models" in the same path of the OSM file loaded in JOSM. In this folder will reside the default SVM models and the models that are being produced by the user from the training process.