# Sparse Personalized Text Generation with Multi-Trajectory Reasoning

Source: [Sparse Personalized Text Generation with Multi-Trajectory Reasoning](<https://www.sota2.com/research/paper/69f1907655a3c32af89fb9b5>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-30T05:00:04.704Z

Authors: Bo Ni, Haowei Fu, Qinwen Ge, Franck Dernoncourt, Samyadeep Basu, Nedim Lipka, Seunghyun Yoon, Yu Wang, Nesreen K. Ahmed, Subhojyoti Mukherjee, Puneet Mathur, Ryan A. Rossi, Tyler Derr

Published: 2026-04-27T20:58:49.000Z

## Abstract

As Large Language Models (LLMs) advance, personalization has become a key mechanism for tailoring outputs to individual user needs. However, most existing methods rely heavily on dense interaction histories, making them ineffective in cold-start scenarios where such data is sparse or unavailable. While external signals (e.g., content of similar users) can offer a potential remedy, leveraging them effectively remains challenging: raw context is often noisy, and existing methods struggle to reason over heterogeneous data sources. To address these issues, we introduce PAT (Personalization with Aligned Trajectories), a reasoning framework for cold-start LLM personalization. PAT first retrieves information along two complementary trajectories: writing-style cues from stylistically similar users and topic-specific context from preference-aligned users. It then employs a reinforcement learning-based, iterative dual-reasoning mechanism that enables the LLM to jointly refine and integrate these signals. Experimental results across real-world personalization benchmarks show that PAT consistently improves generation quality and alignment under sparse-data conditions, establishing a strong solution to the cold-start personalization problem.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Long Text Generation on Stylized Feedback Generation Benchmark](<https://www.sota2.com/research/sota/long-text-generation-on-stylized-feedback-generation-benchmark.md>): Long Text Generation; Stylized Feedback Generation Benchmark; R-1: 24.2.

- [Long Text Generation on Hotel Experience](<https://www.sota2.com/research/sota/long-text-generation-on-hotel-experience.md>): Long Text Generation; Hotel Experience; ROUGE-1: 0.277.

- [Long Text Generation on Amazon Review](<https://www.sota2.com/research/sota/long-text-generation-on-amazon-review.md>): Long Text Generation; Amazon Review; ROUGE-1: 21.

- [Short Text Generation on Stylized Feedback Generation Benchmark](<https://www.sota2.com/research/sota/short-text-generation-on-stylized-feedback-generation-benchmark.md>): Short-text generation; Stylized Feedback Generation Benchmark; ROUGE-1 Score: 0.191.

- [Short Text Generation on Amazon Review](<https://www.sota2.com/research/sota/short-text-generation-on-amazon-review.md>): Short-text generation; Amazon Review; ROUGE-1: 17.7.

- [Long Text Generation on Personalization Benchmark Long Text](<https://www.sota2.com/research/sota/long-text-generation-on-personalization-benchmark-long-text.md>): Long Text Generation; Personalization Benchmark Long Text; R-1 Score: 0.233.

- [Short Text Generation on Personalization Benchmark Short Text](<https://www.sota2.com/research/sota/short-text-generation-on-personalization-benchmark-short-text.md>): Short-text generation; Personalization Benchmark Short Text; R-1: 0.157.

- [Short Text Generation on Hotel Experience](<https://www.sota2.com/research/sota/short-text-generation-on-hotel-experience.md>): Short-text generation; Hotel Experience; ROUGE-1: 13.3.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.24996>)

[PDF](<https://arxiv.org/pdf/2604.24996.pdf>)