INDRA

INDRA World is a knowledge assembly system that ingests causal relations extracted by one or more reading systems from documents, captures these in a standardized representation, and runs an assembly pipeline to recognize full and partial overlaps between relations, filter according to various criteria, and assess belief given the overall support for each relation.

The INDRA World documentation is available here.

Workflows

W2 Reading + integration/assembly

In this workflow INDRA World processes outputs from one or more reading systems (Eidos, Hume or Sofia) as input. We assume that each reading system was already run and produced a number of output files. INDRA World assembly can be done using the command line interface either natively (if an appropriate Python environment and INDRA World and its dependencies are installed) or through Docker.

indra_world --reader-output-files READER_OUTPUT_FILES --ontology-path ONTOLOGY_PATH --output-folder OUTPUT_FOLDER

In the above, READER_OUTPUT_FILES refers to a JSON file with the following structure

{
 "eidos":
 [
 "/path/to/file1",
 "/path/to/file2",
 ],
 "hume": [
 ...
 ]
}

where keys are reading systems and values are lists of paths to their output files.

ONTOLOGY_PATH is a path to a standard ontology YAML file. OUTPUT_FOLDER refers to a folder in which the resulting statements.json JSON-L dump of assembled INDRA Statements will be written. For more optional arguments, see the CLI documentation.

W3 Document management + reading + integration/assembly

In this workflow, INDRA World has access to DART, which keeps track of and provides access to documents, reader outputs, and ontologies through a standardized API. The INDRA World CLI can be parameterized to retrieve reader outputs as well as an ontology through the DART API. An example call to the INDRA World CLI is as follows:

indra_world --reader-output-dart-query READER_OUTPUT_DART_QUERY --ontology-id ONTOLOGY_ID --output-folder OUTPUT_FOLDER

The argument --reader-output-dart-query READER_OUTPUT_DART_QUERY is a path to a JSON file in which the parameters for querying DART are specified following the format of this function. For example,

{
 "readers": ["eidos", "hume"],
 "ontology_id": "49277ea4-7182-46d2-ba4e-87800ee5a315"
}

will query DART for outputs from the eidos and hume readers produced with the ontology ID 49277ea4-7182-46d2-ba4e-87800ee5a315.

Alternatively, if the user wants to hand pick which specific DART reader output records to use for assembly (instead of finding these through a structured query as described above), the --reader-output-dart-keys READER_OUTPUT_DART_KEYS argument can be used which points to a simple text file in which the unique storage keys associated with a list of DART records are given as e.g.,

d6c90753-fe23-43c1-9457-8518b3eaeb65.jsonld
5b901d9a-44b9-4d57-9257-1e0881ed2d06.jsonld
...

In this workflow, the ontology is also registered in DART and can be specified simply through its ID using the --ontology-id argument.

The output folder for INDRA is specified as using --output-folder, as described in W2.

W4 Document management + reading + integration/assembly + HMI

This workflow is different from W3 only in that we need to make sure the INDRA World CLI produces its output in an appropriate folder structure, wiuth output files compatible with Causemos. To achieve this, in addition to the --output-folder argument, we need to supply the --causemos-metadata CAUSEMOS_METADATA argument. Here, CAUSEMOS_METADATA is the path to a JSON file containing metadata that Causemos uses to identify and describe the assembled result. The structure of this JSON file follows the metadata.json entry described here.

An example CLI call for this workflow is as follows: ``` indra_world –reader-output-dart-query READER_OUTPUT_DART_QUERY –ontology-id ONTOLOGY_ID –output-folder OUTPUT_FOLDER –causemos-metadata CAUSEMOS_METADATA

W5 Document management + reading + integration/assembly + HMI + BYOD

This workflow is meaningfully different from W2-4 in that INDRA World here has to be running as a service to support uploading documents in Causemos and performing incremental assembly with respect to an existing Causemos project based on reader output for the new documents.

To run the INDRA World service, follow the instructions here.

Once the service is running, one can build an assembled knowledge base using INDRA World based on reader outputs captured by DART using (1) INDRA World’s Python API by using the Corpus manager, or (2) using the INDRA World assembly dashboard web UI which runs at http://localhost:8001/dashboard under default settings.