Actionable Contextual Explanation System (ACES)

Note: To keep the anonymity of the submission, we can not provide the thing descriptions and the knowledge graphs for simulating the results. Hence, one has to set up the environment and provide appropriate input to the system for generating contextual explanations. To keep the anonymity of the submission, we can not provide the thing descriptions and the knowledge graphs for simulating the results. Hence, one has to set up the environment and provide appropriate input to the system for generating contextual explanations. For testing ACES with exisiting data, please use sensors_data.csv to populate the log data in your environemnt. It is a labeled data set used for demonstraing the prototype in this paper with features (attributes) such as Lamp brighness level(lightPowerStatus), indoor temperature (temperature), humidity, light, Outdoor temperature (Temperature), Outdoor (LightLevel), Brightness status label (lightPowerStatusClass).

1. Setup the context knowledge graph . Setup GraphDB to run SPARQL queries to retrieve the Thing Descriptions GraphDB setup . Upload the knowledge graphs and ontologies to your GraphDB repository. The external ontologies used for ACES are: BRICK Ontology Prov-O Ontology

2. Update the credentials in config.ini (rename the file config.ini.example to config.ini)

3. Install required packages for Python 3.8.2

   python install -r requirements.txt

4. Run the system:

   a. Using the setup:

   cd contextual_explainer/src/
   python main.py

   Example (simplified) of an interaction and the output of the system:

   Contextual Explanation System
   
   ---------Query-------------
   Enter name of the entity: example_lamp_1
   Enter feature name for finding influence: lightPowerStatus
   Enter datetime of query instance: 2022-01-01T00:00:00Z
   
   ---------Context Discovery-
   Enter the ontology prefix used: hsg
   Enter the ontology uri: <http://.../livingcampus#>
   Enter the relationship name to discover the Thing Descriprions: hasLocation
   
   Surrogate model accuracy: 98.93%
   
   Select a counterfactual: 2 (index of the counterfactual, Light brightness level 0.7)
   
   Contextual Explanation:
   Room Temperature has positiveInfluence on light brightness Level of example_lamp_1
   
   The preserved influences between ex:example_lamp_1 and brick:Temperature have been reviewed 16 times by the users as positiveInfluence
   
   Average ratings for the preserved contextual influence:
   {'positiveInfluence': 0.75, 'negativeInfluence': 0.0}
   
   Please provide your feedback for future reference by the users: The relationship looks plausible and it may be because the room was occupied by the employees at the given time.
   Do you agree with the found relationship? yes/no: yes