[Bug]: Invalid label generated for upsert_triplet method when creating a KG for Neptune Analytics

[kevinphillips81]

Bug Description

When generating a KG using the KnowledgeGraphIndex, on occasions the labels extracted from the documents include backticks. Backticks are supported by Neptune, and the code already provisions these within the query. When the label itself includes the backticks, the query fails along with the KG creation.

!pip install llama-index-llms-bedrock llama-index-graph-stores-neptune llama-index-embeddings-bedrock -q

Using version 0.3.0 of llama-index-graph-stores-neptune

Version

0.12.6

Steps to Reproduce

llm = Bedrock(model="anthropic.claude-3-haiku-20240307-v1:0")
embed_model = BedrockEmbedding(model="amazon.titan-embed-text-v1")

Settings.llm = llm
Settings.embed_model = embed_model
Settings.chunk_size = 512

docs = SimpleWebPageReader(
    html_to_text=True,
    metadata_fn=lambda url:{'url': url}
).load_data(web_urls)

graph_store = NeptuneAnalyticsGraphStore(
    graph_identifier=g-abc1d23e45
)

storage_context = StorageContext.from_defaults(graph_store=graph_store)

# NOTE: can take a while!
index = KnowledgeGraphIndex.from_documents(
    docs,
    storage_context=storage_context,
    max_triplets_per_chunk=100,
)

Relevant Logs/Tracbacks

No response

[dosubot]

Hey @kevinphillips81! I'm here to help you with any bugs, questions, or contributions. Let's tackle this together!

To address the issue of invalid labels with backticks being generated by the upsert_triplet method in the llama-index-graph-stores-neptune package, you should ensure that the rel parameter is properly sanitized before being used in the query. The current implementation replaces spaces with underscores and converts the relationship type to uppercase, which is correct. However, if backticks or other invalid characters are present, you should further sanitize the rel string to remove or replace these characters.

Here's a snippet of the upsert_triplet method for reference:

def upsert_triplet(self, subj: str, rel: str, obj: str) -> None:
    """Add triplet to the graph."""
    query = """
        MERGE (n1:`%s` {id:$subj})
        MERGE (n2:`%s` {id:$obj})
        MERGE (n1)-[:`%s`]->(n2)
    """

    prepared_statement = query % (
        self.node_label,
        self.node_label,
        rel.replace(" ", "_").upper(),
    )

    self.query(prepared_statement, {"subj": subj, "obj": obj})

If the issue persists, you may need to implement additional logic to sanitize the rel string, ensuring it conforms to the valid label syntax for your specific use case ^[1].

To continue talking to Dosu, mention @dosu.

---

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

[kevinphillips81]

To mitigate, you need to update the upsert_triplet method in the /graph_stores/neptune/base.py file as follows:

def upsert_triplet(self, subj: str, rel: str, obj: str) -> None:
        """Add triplet to the graph."""
        query = """
            MERGE (n1:`%s` {id:$subj})
            MERGE (n2:`%s` {id:$obj})
            MERGE (n1)-[:`%s`]->(n2)
        """

        prepared_statement = query % (
            self.node_label.replace("`",""),
            self.node_label.replace("`",""),
            rel.replace(" ", "_").replace("`","").upper(),
        )

        self.query(prepared_statement, {"subj": subj.replace("`",""), "obj": obj.replace("`","")})

[Riddhimaan-Senapati]

Created a PR to solve this issue using your implemented solution @kevinphillips81