📊 H2HMem: A Multimodal Memory Benchmark for Agents in Human-Human Interactions

📄 Paper

This dataset is introduced in the following research work:

H2HMem: A Multimodal Memory Benchmark for Agents in Human-Human Interactions

-📄 arXiv: https://arxiv.org/abs/2606.09461v1
-💻 Code: https://github.com/varib1/H2HMEM
-📊 Dataset: https://huggingface.co/datasets/varib/H2HMEM
-🌐 Project Page: https://h2hmemprojectpage.vercel.app/
-🏆 Leaderboard: https://h2hmemleaderboard1.vercel.app/
-📤 Leaderboard Submission: https://huggingface.co/spaces/varib/H2HMEM-Submit

---

H2HMem is a benchmark designed to evaluate multimodal memory and reasoning capabilities in LLM-based agents across dyadic and multi-party human-human conversations.

---

📊 Dataset Overview

---

🏷️ Task Taxonomy

The benchmark includes three major categories of memory evaluation tasks:

🧠 Memory Recall

🧩 Memory Reasoning

⚙️ Memory Application