QA over KGs

Question 1

What are the main stages of the KGQA pipeline?

Answer 1

1) Natural Language Question
2) Query Formulation
3) Structured Query (SPARQL)
4) Query Execution
5) Knowledge Graph Execution
6) Result Processing
7) Answer Generation

Question 2

What occurs in the ‘Query Formulation’ stage of KGQA?

Answer 2

Translating the natural language question into a formal structured query (e.g., SPARQL) by recognizing entities, relations, and constructing the logical form.

Question 3

What is involved in ‘Query Execution’ and ‘Knowledge Graph Execution’?

Answer 3

Running the SPARQL query against the knowledge graph endpoint to retrieve matching triples or subgraphs that answer the query.

Question 4

What does ‘Result Processing’ entail in the KGQA pipeline?

Answer 4

Aggregating raw query results, filtering or deduplicating, ranking candidate answers, and formatting the data for the final generation step.

Question 5

Describe the ‘Answer Generation’ stage in KGQA.

Answer 5

Using the processed results (and optionally retrieved text via RAG) to generate a coherent, natural-language answer for the user through templates or an LLM.

Question 6

Where and how does Retrieval-Augmented Generation (RAG) integrate into the pipeline?

Answer 6

Between Result Processing and Answer Generation: format retrieved KG data as context, optionally augment with text or embeddings, then feed to an LLM to produce the final answer.

Question 7

Name the four key challenges highlighted on the KGQA pipeline slide.

Answer 7

1) Entity Recognition
2) Relation Mapping
3) Query Construction
4) Answer Ranking

Question 8

What issues arise in ‘Entity Recognition’ within KGQA?

Answer 8

Identifying and disambiguating entity mentions in user questions, handling synonyms, polysemy, and acronyms.

Question 9

What is the ‘Relation Mapping’ challenge in KGQA?

Answer 9

Mapping natural-language predicates to the correct KG properties, dealing with varied phrasing and implicit relations.

Question 10

What does the ‘Query Construction’ challenge refer to?

Answer 10

Assembling the recognized entities and relations into a valid SPARQL query, capturing the intended logic and filters.

Question 11

Explain the ‘Answer Ranking’ challenge in KGQA.

Answer 11

Ordering multiple candidate answers by relevance or correctness, selecting the best for presentation to the user.