[Question]: Problems with entity extractor

[JLombar]

Question Validation

• I have searched both the documentation and discord for an answer.

Question

I am using an entity extractor with my ingestion pipeline but the entities do not appear in the metadata:

self.pipeline = IngestionPipeline(
      transformations=[
            IngestionService.TextCleaner(),
            self.text_splitter,
            self.embed_model,
            TitleExtractor(),
            EntityExtractor(prediction_threshold=0.5),
        ],
      docstore=MongoDocumentStore.from_uri(uri=MONGO_URI),
      vector_store=self.vector_store,
      cache=self.ingest_cache,
      docstore_strategy=DocstoreStrategy.UPSERTS_AND_DELETE,
      disable_cache=False
  )

nodes = await self.pipeline.arun(documents=documents, show_progress=True)

for node in nodes:
            print(node.metadata)

This is the result of the print:

{'page_label': '1', 'file_name': 'nazaries_USO-CORRECTO-DEL-AIRE-ACONDICIONADO-E-ILUMINACIÓN-EN-OFICINA.pdf', 'file_path': '/home/jlombar/ai-llamaindex-agent/data/nazaries_USO-CORRECTO-DEL-AIRE-ACONDICIONADO-E-ILUMINACIÓN-EN-OFICINA.pdf', 'file_type': 'application/pdf', 'file_size': 174299, 'creation_date': '2025-04-29', 'last_modified_date': '2025-04-29', 'document_title': ' "Office Climate Control and Energy-Efficient Lighting Guidelines: Simultaneous Lighting at Workplace for Uniform Illumination and Environmental Respect"\n\nThis title effectively communicates the main topics of the document while also emphasizing the importance of energy conservation, environmental respect, and uniform illumination in office settings.', 'urls': {}, 'annotated_text': 'ser encendidas simultáneamente al inicio de la jornada laboral para asegurar una iluminación uniforme.'}

Is there something I need to change in the pipeline? Or what is the error?

[dosubot]

Here are a few things you can check to troubleshoot why the EntityExtractor is not adding entities to the metadata:

1. Model Configuration: Ensure that the EntityExtractor is correctly configured with the appropriate model. The default model is "tomaarsen/span-marker-mbert-base-multinerd". Verify that this model is suitable for your data and is correctly loaded.

2. Prediction Threshold: The prediction_threshold parameter is set to 0.5, which means entities are only extracted if their prediction confidence is at least 0.5. You might want to experiment with lowering this threshold to see if it affects the extraction results.

3. Order of Transformations: The order of transformations in the pipeline can affect the outcome. Ensure that the EntityExtractor is placed correctly in the sequence, especially after any text cleaning or splitting operations that might alter the text structure.

4. Device Configuration: Check if the device parameter is set correctly, especially if you are using GPU for extraction. Incorrect device settings might lead to extraction issues.

5. Entity Map: If you have a specific entity map, ensure it is correctly defined. The default entity map is empty, but you can customize it to map entity class names to usable names.

6. Debugging Information: Although specific logging or debugging information wasn't found, you might want to add logging around the entity extraction process to see if entities are being detected but not added to metadata.

If none of these steps resolve the issue, consider reviewing the document content to ensure it contains recognizable entities according to the model used. Additionally, verify that the EntityExtractor is correctly integrated into the pipeline without any configuration errors.

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other}

[logan-markewich]

@JLombar judging by the file name, im guessing your file is non-english? You might want to change the entity model to something multilingual or lower the threshold

[dosubot]

Hi, @JLombar. I'm Dosu, and I'm helping the LlamaIndex team manage their backlog and am marking this issue as stale.

Issue Summary:

• You reported that entities extracted by the EntityExtractor do not appear in the node metadata.
• Suggestions included checking model configuration, prediction threshold, transformation order, device settings, and entity map.
• I recommended adding logging for debugging purposes.
• A possible language mismatch was noted, with advice to use a multilingual entity model or lower the threshold.
• No confirmed resolution has been provided yet.

Next Steps:

• Please let me know if this issue is still relevant with the latest version of LlamaIndex by commenting here.
• Otherwise, this issue will be automatically closed in 7 days.

Thank you for your understanding and contribution!