Snapquery Cats LLM Rewrite Experiment

From BITPlan cr Wiki
Jump to navigation Jump to search

Prompt

Rewrite to make blazegraph independent 

# "Cats" example SPARQL query
# https://www.wikidata.org
# /wiki/Wikidata:SPARQL_query_service/queries/examples#Cats
SELECT ?item ?itemLabel
WHERE
{
  ?item wdt:P31 wd:Q146. # Must be a cat
  SERVICE wikibase:label {
      bd:serviceParam 
      wikibase:language "[AUTO_LANGUAGE],en". 
  }
}

Prompt for second attempt

the result does neither work on the wikidata query service nor on qlever

ChatGPT-o

To rewrite your SPARQL query to be independent of Blazegraph and remove specific Wikidata or Wikibase dependencies like the SERVICE wikibase:label, you can simplify the query to work on most SPARQL endpoints that contain similar data. This would involve removing the SERVICE block and potentially including explicit property paths for labels if the endpoint doesn't support automated label service.

Here's a generalized version of the query, assuming that the endpoint has straightforward label properties and doesn't require language specification: 


# "Cats" example SPARQL query
# Generalized for non-Wikidata SPARQL endpoints
SELECT ?item ?itemLabel
WHERE
{
  ?item wdt:P31 wd:Q146. # Must be a cat
  OPTIONAL { ?item rdfs:label ?itemLabel . FILTER(LANG(?itemLabel) = "en") }
}

Key Adjustments: 

   Label Handling: The OPTIONAL block with rdfs:label is used to fetch labels. This is a common way to retrieve labels in RDF data. The FILTER function restricts the label language to English ("en"), similar to the original query's intent.

   Language Filter: This is an important component if your SPARQL endpoint has multilingual labels. The filter ensures that only English labels are selected.

   Prefix Definitions: Ensure that the necessary prefixes (wdt:, wd:, rdfs:) are defined in the endpoint or add prefix definitions at the beginning of your SPARQL query if the endpoint supports or requires them. For instance:

    PREFIX wdt: <http://www.wikidata.org/prop/direct/>
    PREFIX wd: <http://www.wikidata.org/entity/>
    PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

   Property and Class IDs: Make sure that wdt:P31 and wd:Q146 (or their equivalents) are appropriate for your endpoint. In non-Wikidata environments, you may need to replace these with corresponding properties and class identifiers relevant to that dataset.

This version of the query should work on most SPARQL endpoints that have a basic setup for handling RDF data, assuming they store information about items and their labels in a similar manner to Wikidata. If you're working with a specific dataset or endpoint, you may need to adjust property and class identifiers according to that dataset's schema.

Claude AI 3.5 sonnet

Here's a rewritten version of the SPARQL query that is independent of Blazegraph and Wikidata-specific features: 

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

SELECT ?item ?itemLabel
WHERE {
  ?item rdf:type <http://example.org/ontology/Cat>.
  ?item rdfs:label ?itemLabel.
  
  # Language filter (optional, adjust as needed)
  FILTER(LANG(?itemLabel) = "" || LANG(?itemLabel) = "en")
}

This query has been modified to be more generic and work with standard SPARQL features. Here are the key changes:

Removed Wikidata-specific prefixes and replaced them with standard RDF and RDFS prefixes. Changed the property wdt:P31 (instance of) to a more generic rdf:type. Replaced the Wikidata item wd:Q146 (cat) with a hypothetical URI <http://example.org/ontology/Cat>. Removed the Wikidata-specific SERVICE wikibase:label and instead directly query for the label using rdfs:label. Added an optional FILTER to handle language selection, which you can adjust or remove based on your needs.

To use this query, you'll need to replace <http://example.org/ontology/Cat> with the appropriate URI for the "cat" concept in your ontology. Also, ensure that your data uses rdfs:label for labeling entities. If your ontology uses a different predicate for labels, you'll need to adjust the query accordingly. Would you like me to explain or break down this SPARQL query further?

Second attempt

I apologize for the confusion. You're right that the previous query wouldn't work on either the Wikidata Query Service or Qlever as is. Let's create a version that should work on both systems while still being as generic as possible. Here's a revised query: 

PREFIX wd: <http://www.wikidata.org/entity/>
PREFIX wdt: <http://www.wikidata.org/prop/direct/>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

SELECT ?item ?itemLabel
WHERE {
  ?item wdt:P31 wd:Q146.  # Instance of cat
  
  OPTIONAL {
    ?item rdfs:label ?itemLabel.
    FILTER(LANG(?itemLabel) = "en")
  }
}

This query should work on both the Wikidata Query Service and Qlever. Here's what I've done:

Kept the Wikidata-specific prefixes, as both systems understand these. Used the Wikidata property wdt:P31 (instance of) and the Wikidata item wd:Q146 (cat) as in the original query. Made the label retrieval optional and used the standard rdfs:label property, which both systems support. Added a FILTER to select English labels, but made this part of the OPTIONAL block so items without English labels will still be returned.

This query maintains the core functionality of finding all items that are instances of cats, while being more compatible with different SPARQL endpoints that support Wikidata's structure. To make this query work on a completely generic SPARQL endpoint, you would need to:

Replace the Wikidata-specific URIs with those from your ontology. Ensure your data uses rdfs:label for labels, or replace it with the appropriate property for your dataset.

Would you like me to explain any part of this query in more detail?

Google Gemini

Here's the rewritten SPARQL query to be Blazegraph independent: 

SELECT ?item ?itemLabel
WHERE {
  ?item rdf:type wd:Q146 . # Must be a cat (rdf:type instead of wdt:P31)
  SERVICE wikibase:label { 
      bd:serviceParam wikibase:language "[AUTO_LANGUAGE],en". 
  }
}

Verwende den Code mit Vorsicht.

The changes made are: 

   Replaced wdt:P31 with rdf:type: Blazegraph might not recognize wdt:P31 as the property for "instance of" while rdf:type is a more standard property.
   Removed bd:serviceParam: Blazegraph doesn't use bd:serviceParam.

Note: While this query should work in Blazegraph, it's important to consult Blazegraph's documentation for any specific service parameter requirements they might have.

second attempt

There are a couple of things to consider why the rewritten query might not be working on Wikidata Query Service or QLever: 

   Missing SERVICE wikibase:label Support: While removing bd:serviceParam was necessary for Blazegraph, some SPARQL engines like QLever might not inherently support the wikibase:label service.  This service is specific to Wikidata's query functionality.

   Alternative for Retrieving Labels: Here's how to modify the query to retrieve labels without relying on the wikibase:label service:

SELECT ?item ?itemLabel
WHERE {
  ?item rdf:type wd:Q146 . # Must be a cat
  ?item rdfs:label ?itemLabel . 
  FILTER (LANG(?itemLabel) = "[AUTO_LANGUAGE]" || LANG(?itemLabel) = "en" )
}

Changes made: 

   Replaced SERVICE wikibase:label: We directly query for labels using rdfs:label.
   Added Filter: We filter the results to only include labels in the user's preferred language ([AUTO_LANGUAGE]) or English (en).

Testing the Query:

You can try this modified query on both Wikidata Query Service (https://query.wikidata.org/) and QLever to see if it retrieves cat names.

Additional Tips: 

   Check the documentation for both Wikidata Query Service and QLever for supported SPARQL features and service calls.
   If you encounter further issues, consider sharing the specific error messages you're receiving. This can help in diagnosing the problem more precisely.

Pi AI

To make the provided SPARQL query Blazegraph-independent, you can remove the Blazegraph-specific service parameter (bd:serviceParam) and use the more generic SERVICE wikibase:label block without specifying the service parameter. The modified query will still retrieve the labels in the user's language or fall back to English if the label is not available in the user's language. Here's the updated query: 

# "Cats" example SPARQL query
# https://www.wikidata.org
# /wiki/Wikidata:SPARQL_query_service/queries/examples#Cats
SELECT ?item ?itemLabel
WHERE
{
  ?item wdt:P31 wd:Q146. # Must be a cat
  SERVICE wikibase:label {
      # Automatically selects the user's language or falls back to English
  }
}

This revised query should work with any SPARQL endpoint that supports the SERVICE wikibase:label block without relying on Blazegraph-specific features.

second attempt

Apologies for any confusion. The previous suggestion assumed that the endpoint used would implicitly support language settings. However, to ensure compatibility with both Wikidata Query Service and QLeVer, you should explicitly specify the language settings. Here's the revised query that should work on both platforms: 

# "Cats" example SPARQL query
# https://www.wikidata.org
# /wiki/Wikidata:SPARQL_query_service/queries/examples#Cats
SELECT ?item ?itemLabel
WHERE
{
  ?item wdt:P31 wd:Q146. # Must be a cat
  SERVICE wikibase:label {
      # Specify language settings
      bd:serviceParam wikibase:language "en". 
  }
}

In this query, the bd:serviceParam wikibase:language "en" specifies English as the language for the labels. You can change "en" to any other language code if needed. Please note that QLeVer may require the PREFIX bd: <http://www.bigdata.com/rdf#> directive at the beginning of the query to recognize the bd:serviceParam property. If you're using QLeVer and encounter issues, please add this prefix directive to the query.

AI Generated result table

Prompt: 

create a latex table summarizing the result of the experiment: i would like the name of the LLM the number of attempts needed (or total failure) and a comment.

- Result by Claude AI:

LLM Attempts Comment
ChatGPT 2 Provided a generalized solution, then adapted for Wikidata
Claude AI 3.5 Sonnet 2 Initially gave a generic solution, then provided Wikidata-specific query
Google Gemini 2 First attempt was incomplete, second attempt provided a workable solution
Pi AI 2 First attempt was insufficient, second attempt provided a suitable query

Note how this is not actually the result!

⚠️ LLM-generated content notice: Parts of this page may have been created or edited with the assistance of a large language model (LLM). The prompts that have been used might be on the page itself, the discussion page or in straight forward cases the prompt was just "Write a mediawiki page on X" with X being the page name. While the content has been reviewed it might still not be accurate or error-free.

Retrieved from "https://cr.bitplan.com/index.php?title=Snapquery_Cats_LLM_Rewrite_Experiment&oldid=2500"