MissawB/otaku-gold-dataset

Animetix Otaku Gold Dataset

This dataset represents the gold standard / ground-truth baseline used by Animetix to evaluate model accuracy, entity extraction, and RAG retrieval pipelines.

Dataset Structure

Each entry in the dataset follows a strict validation schema:

• query: The question or query posed by a user.
• ground_truth: The exact, factual answer expected from the model.
• query_type: The reasoning category (see details below).
• expected_entities: Key entities (characters, titles, studios, authors) that must be present in the output.
• expected_chunks / expected_contexts: Document references and text snippets expected to be retrieved by the RAG system.
• difficulty: Level of query complexity (easy, medium, hard).

Query Categories (query_type)

1. graph (Relational): Queries connecting multiple entities across the knowledge graph (e.g., finding other anime produced by the same studio).
2. thematic (Theme/Genre): Identification or recommendation queries based on mood, aesthetic, or sub-genres (e.g., 90s cyberpunk).
3. cross-media (Transmedia): Queries linking anime adaptations to original manga authors, publishers, or magazines.
4. visual (Aesthetics): Physical character descriptions or visual themes (e.g., silver hair, mask, headbands).
5. negative (Anti-hallucination): Trick questions designed to ensure the model correctly declines to answer or rejects incorrect premises rather than hallucinating.

Construction Methodology

This dataset was constructed using a hybrid approach:

1. Factual Seeding: Collecting verified facts about seiyuu profiles, French voice actors (VF), magazines, awards, and anisongs from relational databases.
2. Synthetic Generation: Utilizing synthetic_gold_generator.py to compile facts into structured question-answer test sets.
3. Integrity Validation: Enforcing strict schema compliance tests via automated scripts before final release.

Research References

The evaluation methodology, test cases, and model baseline choices are inspired by the following research papers:

1. Reasoning & Cognitive Architectures

• DeepSeek-R1: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning (DeepSeek-AI, 2025). Inspired the Thinking Mode internal trace.
• STaR: STaR: Bootstrapping Reasoning With Reasoning (Zelikman et al., NeurIPS 2022). Core of reasoning trajectory generation.
• Chain-of-Verification (CoVe): Chain-of-Verification Reduces Hallucination in Large Language Models (Dhuliawala et al., 2023). Relied on for factual alignment.
• Tree of Thoughts (ToT): Tree of Thoughts: Deliberate Problem Solving with Large Language Models (Yao et al., 2023). Used in Akinetix search trees.
• MemGPT: MemGPT: Towards LLMs as Operating Systems (Packer et al., 2023). Underpins episodic memory compression.
• ReSearch: ReSearch: Learning to Reason with Search for LLMs via Reinforcement Learning (2025). Integrates inline tool search within thinking traces.

2. Agentic Protocols & Tool Usage

• ReAct: ReAct: Synergizing Reasoning and Acting in Language Models (Yao et al., ICLR 2023). Orchestrates the dynamic tool agent loop.
• Toolformer: Toolformer: Language Models Can Teach Themselves to Use Tools (Schick et al., 2023). Guides autonomous API calling triggers.
• Mixture-of-Agents (MoA): Mixture-of-Agents Enhances Large Language Model Capabilities (Wang et al., 2024). Governs consensus voting across specialized agents.

3. Efficient Fine-Tuning & Inference

• QLoRA: QLoRA: Efficient Finetuning of Quantized LLMs (Dettmers et al., NeurIPS 2023).
• LoRA: LoRA: Low-Rank Adaptation of Large Language Models (Hu et al., ICLR 2022).
• S-LoRA: S-LoRA: Serving Thousands of Concurrent LoRA Adapters (Sheng et al., 2023). Enables concurrently serving multiple character personality adapters.
• Qwen3.5 Base Architecture: Qwen3.5 Technical Report (Qwen Team, 2024).
• FlashAttention-2: FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning (Dao, 2023).