Excited to share our work on <strong>Parametric Social Identity Injection (PSII)</strong> for socially diverse public opinion simulation with LLM agents.</p>\n<p>We identify <strong>Diversity Collapse</strong>, a phenomenon where prompt-conditioned LLM agents produce overly homogeneous responses and flattened inter-group differences. PSII addresses this by injecting explicit demographic and value-orientation vectors into intermediate hidden states, enabling fine-grained and controllable identity modulation beyond prompt-only personas.</p>\n<p>On World Values Survey questions, PSII improves distributional fidelity and diversity across multiple open-source LLMs.</p>\n<p>Demo: <a href=\"https://psii-demo.streamlit.app\" rel=\"nofollow\">https://psii-demo.streamlit.app</a><br>Code: <a href=\"https://github.com/halsayxi/PSII\" rel=\"nofollow\">https://github.com/halsayxi/PSII</a><br>Paper: <a href=\"https://arxiv.org/abs/2603.16142\" rel=\"nofollow\">https://arxiv.org/abs/2603.16142</a><br>Model: <a href=\"https://huggingface.co/hexi222/psii-identity-representations\">https://huggingface.co/hexi222/psii-identity-representations</a></p>\n<p><a href=\"https://cdn-uploads.huggingface.co/production/uploads/676aff243d3de6571fbdda8e/teZYYDbn4GXyr6x7aKyv3.png\" rel=\"nofollow\"><img src=\"https://cdn-uploads.huggingface.co/production/uploads/676aff243d3de6571fbdda8e/teZYYDbn4GXyr6x7aKyv3.png\" alt=\"framework\"></a></p>\n","updatedAt":"2026-06-08T03:19:20.027Z","author":{"_id":"676aff243d3de6571fbdda8e","avatarUrl":"/avatars/397353cabdfea76c4cfab2d1d2f37247.svg","fullname":"Wang Hexi","name":"hexi222","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6840781569480896},"editors":["hexi222"],"editorAvatarUrls":["/avatars/397353cabdfea76c4cfab2d1d2f37247.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2603.16142","authors":[{"_id":"6a228191047f837f98677846","user":{"_id":"676aff243d3de6571fbdda8e","avatarUrl":"/avatars/397353cabdfea76c4cfab2d1d2f37247.svg","isPro":false,"fullname":"Wang Hexi","user":"hexi222","type":"user","name":"hexi222"},"name":"Hexi Wang","status":"claimed_verified","statusLastChangedAt":"2026-06-05T15:05:59.841Z","hidden":false},{"_id":"6a228191047f837f98677847","name":"Yujia Zhou","hidden":false},{"_id":"6a228191047f837f98677848","name":"Bangde Du","hidden":false},{"_id":"6a228191047f837f98677849","name":"Qingyao Ai","hidden":false},{"_id":"6a228191047f837f9867784a","name":"Yiqun Liu","hidden":false}],"publishedAt":"2026-06-01T00:00:00.000Z","submittedOnDailyAt":"2026-06-08T00:00:00.000Z","title":"Parametric Social Identity Injection and Diversification in Public Opinion Simulation","submittedOnDailyBy":{"_id":"676aff243d3de6571fbdda8e","avatarUrl":"/avatars/397353cabdfea76c4cfab2d1d2f37247.svg","isPro":false,"fullname":"Wang Hexi","user":"hexi222","type":"user","name":"hexi222"},"summary":"Large language models (LLMs) have recently been adopted as synthetic agents for public opinion simulation, offering a promising alternative to costly and slow human surveys. Despite their scalability, current LLM-based simulation methods fail to capture social diversity, producing flattened inter-group differences and overly homogeneous responses across demographic groups. We identify this limitation as a Diversity Collapse phenomenon in LLM hidden representations, where distinct social identities become increasingly indistinguishable across layers. Motivated by this observation, we propose Parametric Social Identity Injection (PSII), a general framework that injects explicit, parametric representations of demographic attributes and value orientations directly into intermediate hidden states of LLMs. Unlike prompt-based persona conditioning, PSII enables fine-grained and controllable identity modulation at the representation level. Extensive experiments on the World Values Survey using multiple open-source LLMs show that PSII significantly improves distributional fidelity and diversity, reducing KL divergence to real-world survey data while enhancing overall diversity. This work provides new insights into representation-level control of LLM agents and advances scalable, diversity-aware public opinion simulation.","upvotes":0,"discussionId":"6a228191047f837f9867784b","projectPage":"https://psii-demo.streamlit.app/","githubRepo":"https://github.com/halsayxi/PSII","githubRepoAddedBy":"user","ai_summary":"Large language models suffer from reduced social diversity in public opinion simulation due to identity indistinction in hidden representations, which is addressed through a parametric injection framework that enhances demographic representation fidelity and diversity.","ai_keywords":["large language models","public opinion simulation","diversity collapse","hidden representations","parametric social identity injection","demographic attributes","value orientations","representation-level control","KL divergence","World Values Survey"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":2,"organization":{"_id":"628735cbc83a2d6ab8d14a66","name":"Tsinghua","fullname":"Tsinghua University","avatar":"https://www.gravatar.com/avatar/6c5c1441e3283e7543342e59277ea219?d=retro&size=100"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[],"acceptLanguages":["en"],"organization":{"_id":"628735cbc83a2d6ab8d14a66","name":"Tsinghua","fullname":"Tsinghua University","avatar":"https://www.gravatar.com/avatar/6c5c1441e3283e7543342e59277ea219?d=retro&size=100"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2603/2603.16142.md"}">
Parametric Social Identity Injection and Diversification in Public Opinion Simulation
Abstract
Large language models suffer from reduced social diversity in public opinion simulation due to identity indistinction in hidden representations, which is addressed through a parametric injection framework that enhances demographic representation fidelity and diversity.
Large language models (LLMs) have recently been adopted as synthetic agents for public opinion simulation, offering a promising alternative to costly and slow human surveys. Despite their scalability, current LLM-based simulation methods fail to capture social diversity, producing flattened inter-group differences and overly homogeneous responses across demographic groups. We identify this limitation as a Diversity Collapse phenomenon in LLM hidden representations, where distinct social identities become increasingly indistinguishable across layers. Motivated by this observation, we propose Parametric Social Identity Injection (PSII), a general framework that injects explicit, parametric representations of demographic attributes and value orientations directly into intermediate hidden states of LLMs. Unlike prompt-based persona conditioning, PSII enables fine-grained and controllable identity modulation at the representation level. Extensive experiments on the World Values Survey using multiple open-source LLMs show that PSII significantly improves distributional fidelity and diversity, reducing KL divergence to real-world survey data while enhancing overall diversity. This work provides new insights into representation-level control of LLM agents and advances scalable, diversity-aware public opinion simulation.
Community
Excited to share our work on Parametric Social Identity Injection (PSII) for socially diverse public opinion simulation with LLM agents.
We identify Diversity Collapse, a phenomenon where prompt-conditioned LLM agents produce overly homogeneous responses and flattened inter-group differences. PSII addresses this by injecting explicit demographic and value-orientation vectors into intermediate hidden states, enabling fine-grained and controllable identity modulation beyond prompt-only personas.
On World Values Survey questions, PSII improves distributional fidelity and diversity across multiple open-source LLMs.
Demo: https://psii-demo.streamlit.app
Code: https://github.com/halsayxi/PSII
Paper: https://arxiv.org/abs/2603.16142
Model: https://huggingface.co/hexi222/psii-identity-representations
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2603.16142 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2603.16142 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.