## Flowchart: Claim Verification System
### Overview
The diagram illustrates a technical pipeline for claim verification using natural language processing (NLP) and knowledge graph (KG) integration. It demonstrates how an input text is processed through preprocessing, knowledge graph analysis, and a fine-tuned language model to generate structured outputs including claim validation, rationale, and confidence scores.
### Components/Axes
1. **Input Text**:
- Example: "Steven Tyler has never been a part of the band Aerosmith."
2. **Preprocessing**:
- Sub-components: NER (Named Entity Recognition), Coreference Resolution, KG Entity Linking, Compartmentalization.
3. **Knowledge Graph (KG)**:
- Visualized as interconnected nodes (entities) and edges (relationships).
4. **KG Triple Retrieval Algorithm**:
- Processes KG data to extract relevant triplets.
5. **Finetuned ClaimVer LLM**:
- A language model specialized for claim verification.
6. **Outputs**:
- Structured results including Claim, Prediction, Relevant Triplets & TMS, Rationale, and Score (KAS).
### Detailed Analysis
- **Input Text**: The example claim states "Steven Tyler has never been a part of the band Aerosmith."
- **Preprocessing**:
- NER identifies entities (e.g., "Steven Tyler," "Aerosmith").
- Coreference resolution links pronouns to entities.
- KG Entity Linking maps entities to KG concepts.
- Compartmentalization organizes data into structured units.
- **KG Triple Retrieval Algorithm**:
- Extracts triplets like `[('Aerosmith', 'has part(s)', 'Steven Tyler')]` with confidence score `1.0`.
- **Finetuned ClaimVer LLM**:
- Combines retrieved triplets with input text to generate outputs.
- **Outputs**:
- **Claim**: Repeats the input claim.
- **Prediction**: "Contradictory" (indicating the claim is false).
- **Relevant Triplets & TMS**: Highlights the triplet `[('Aerosmith', 'has part(s)', 'Steven Tyler')]`.
- **Rationale**: Explains that the triplet establishes a relationship between Steven Tyler and Aerosmith, refuting the claim.
- **Score (KAS)**: `0.047` (low score indicates high confidence in the claim's falsity).
### Key Observations
1. The system identifies a contradiction between the input claim and KG data.
2. The retrieved triplet `[('Aerosmith', 'has part(s)', 'Steven Tyler')]` directly challenges the claim.
3. The low KAS score (`0.047`) suggests strong evidence against the claim's validity.
4. The rationale explicitly links the triplet to the claim's contradiction.
### Interpretation
This flowchart represents a hybrid NLP-KG system for fact-checking. The low KAS score (`0.047`) indicates the model is highly confident the claim is false, supported by the KG triplet showing Steven Tyler's association with Aerosmith. The "Contradictory" prediction aligns with the rationale, demonstrating how KG data and language models collaborate to validate claims. The system's design emphasizes transparency by providing both numerical scores and human-readable explanations (rationale). The example highlights the pipeline's ability to detect factual inconsistencies in claims about entity relationships.
