sz_semantics 1.3.3

Transform JSON output from Senzing SDK for use with graph technologies, semantics, and downstream LLM integrations

sz_semantics

Transform JSON output from the Senzing SDK for use with graph technologies, semantics, and downstream LLM integration.

Install

This library uses poetry for demos:

poetry update

Otherwise, to use the library:

pip install sz_sematics

For the gRCP server, if you don't already have Senzing and its gRPC server otherwise installed pull the latest Docker container:

docker pull senzing/serve-grpc:latest

Usage: Masking PII

Mask the PII values within Senzing JSON output with tokens which can be substituted back later. For example, mask PII values before calling a remote service (such as an LLM-based chat) then unmask returned text after the roundtrip, to maintain data privacy.

import json
from sz_semantics import Mask

data: dict = { "ENTITY_NAME": "Robert Smith" }

sz_mask: Mask = Mask()
masked_data: dict = sz_mask.mask_data(data)

masked_text: str = json.dumps(masked_data)
print(masked_text)

unmasked: str = sz_mask.unmask_text(masked_text)
print(unmasked)

For an example, run the demo1.py script with a data file which captures Senzing JSON output:

poetry run python3 demo1.py data/get.json

The two lists Mask.KNOWN_KEYS and Mask.MASKED_KEYS enumerate respectively the:

• keys for known elements which do not require masking
• keys for PII elements which require masking

Any other keys encountered will be masked by default and reported as warnings in the logging. Adjust these lists as needed for a given use case.

For work with large numbers of entities, subclass KeyValueStore to provide a distributed key/value store (other than the Python built-in dict default) to use for scale-out.

Usage: gRPC Client/Server

To use SzClient to simplify access to the Senzing SDK, first launch the serve-grpc container and run it in the background:

docker run -it --publish 8261:8261 --rm senzing/serve-grpc

For example code which runs entity resolution on the "truthset" collection of datasets:

import pathlib
import tomllib
from sz_semantics import SzClient

with open(pathlib.Path("config.toml"), mode = "rb") as fp:
    config: dict = tomllib.load(fp)

data_sources: typing.Dict[ str, str ] = {
    "CUSTOMERS": "data/truth/customers.json",
    "WATCHLIST": "data/truth/watchlist.json",
    "REFERENCE": "data/truth/reference.json",
}

sz: SzClient = SzClient(config, data_sources)
sz.entity_resolution(data_sources)

for ent_json in sz.sz_engine.export_json_entity_report_iterator():
    print(ent_json)

For a demo of running entity resolution on the "truthset", run the demo2.py script:

poetry run python3 demo2.py

This produces the export.json file which is JSONL representing the results of a "get entity" call on each resolved entity.

Note: to show the redo processing, be sure to restart the container each time before re-running the demo2.py script -- although the entity resolution results will be the same even without a container restart.

Usage: Semantic Represenation

Starting with a small SKOS-based taxonomy in the domain.ttl file, parse the Senzing entity resolution (ER) results to generate an RDFlib semantic graph.

In other words, generate the "backbone" for constructing an Entity Resolved Knowledge Graph, as a core componet of a semantic layer.

The example code below serializes the thesaurus generated from Senzing ER results as "thesaurus.ttl" combined with the Senzing taxonomy definitions, which can be used for constructing knowledge graphs:

import pathlib
from sz_semantics import Thesaurus

thesaurus: Thesaurus = Thesaurus()
thesaurus.load_source(Thesaurus.DOMAIN_TTL)

export_path: pathlib.Path = pathlib.Path("data/truth/export.json")

with open(export_path, "r", encoding = "utf-8") as fp_json:
    for line in fp_json:
        for rdf_frag in thesaurus.parse_iter(line, language = "en"):
            thesaurus.load_source_text(
                Thesaurus.RDF_PREAMBLE + rdf_frag,
                format = "turtle",
            )

thesaurus_path: pathlib.Path = pathlib.Path("thesaurus.ttl")
thesaurus.save_source(thesaurus_path, format = "turtle")

For an example, run the demo3.py script to process the JSON file data/truth/export.json which captures Senzing ER exported results:

poetry run python3 demo3.py data/truth/export.json

Check the resulting RDF definitions in the generated thesaurus.ttl file.

---

---

License and Copyright

Source code for sz_semantics plus any logo, documentation, and examples have an MIT license which is succinct and simplifies use in commercial applications.

All materials herein are Copyright © 2025 Senzing, Inc.

Kudos to @brianmacy, @jbutcher21, @docktermj, @cj2001, @jesstalisman-ia, and the kind folks at GraphGeeks for their support.

Star History