Skip to content

Latest commit

 

History

History

stlarm

README.md

Pipeline

Data pipeline

Semantic Trajectories

All data manipulation is done in data.py using the TrajData class and its subclasses.

TrajData can be extended to support different data sources and enrich trajectory data with different semantics. All manipulations are heavily based in pandas' DataFrames.

For example, FoursquareData class uses trajminer lib to download Foursquare data and apply the following steps:

  • Filtering.
  • Semantic enrichment of:
    • Trajectories.
    • Check-ins.

Ontology-based representation

All ontology representation manipulation is done in onto.py (and the utility module onto_mapping.py, which should not need to be used by user code).

We use owlready2 to programmatically manipulate ontologies.

The output of the previous pipeline step can be converted to:

  • STEP representation.
STEP ontology diagram STEP ontology diagram.

Application-specific representation

STEP can be dynamically converted to our proposed application-specific representation.

Note that ./scripts/build_data.py contains code that encapsulates all data building and representation steps (the first 3 steps of our pipeline).

Pseudocode for converting from STEP

We use a systematic approach for converting the STEP representation to our Application-specific one. Algorithm 1 below shows the main procedure. We denote our representation by the converted: prefix.

Algorithm 1

Algorithm 2 shows the auxiliary procedure for associating a Semantic Description to an entity using a relation.

Algorithm 2

Algorithm 3 shows how we process an Episode X's Contextual Elements (CEs). Note that there are three different behaviors:

  1. If the CE is connected to an Episode Y (through an FOI entity), we directly connect the episodes X and Y;
  2. Otherwise (i.e., the CE is connected to Semantic Descriptions, though an FOI entity):
    1. If the CE is a POI, then we create a POI entity and connect the semantic descriptions to it;
    2. Otherwise, we directly connect the episode X to the Semantic Descriptions.

Algorithm 3

Mined Rules

AMIE 3 is available as a JAR package. It is invoked from the terminal to mine rules.

analyzer.py contains code to process the mined rules file and to generate multiple related-analysis.

Metarules

AMIE 3 JAR can be directly used to generate a mapping from rules to metarules. Nonetheless, analyzer.py encapsulates this process and presents a better-formatted version (based in pandas.DataFrame).

metarule_viz.py contains code for visualizing metarules and examples.