Recent advances in Large Language Models (LLMs) have enhanced text-based recommendation by enriching traditional ID-based methods with semantic generalization capabilities. Text-based methods typically encode item textual information via prompt design and generate discrete semantic IDs through item tokenization. However, in domain-specific tasks such as local-life services, simply injecting location information into prompts fails to capture fine-grained spatial characteristics and real-world distance awareness among items. To address this, we propose LGSID, an LLM-Aligned Geographic Item Tokenization Framework for Local-life Recommendation. This framework consists of two key components: (1) RL-based Geographic LLM Alignment, and (2) Hierarchical Geographic Item Tokenization. In the RL-based alignment module, we initially train a list-wise reward model to capture real-world spatial relationships among items. We then introduce a novel G-DPO algorithm that uses pre-trained reward model to inject generalized spatial knowledge and collaborative signals into LLMs while preserving their semantic understanding. Furthermore, we propose a hierarchical geographic item tokenization strategy, where primary tokens are derived from discrete spatial and content attributes, and residual tokens are refined using the aligned LLM’s geographic representation vectors. Extensive experiments on real-world Kuaishou industry datasets show that LGSID consistently outperforms state-of-the-art discriminative and generative recommendation models. Ablation studies, visualizations, and case studies further validate its effectiveness.

Large language models (LLMs) have shown impressive capabilities across various natural language tasks. However, evaluating their alignment with human preferences remains a challenge. To this end, we propose a comprehensive human evaluation framework to assess LLMs’ proficiency in following instructions on diverse real-world tasks. We construct a hierarchical task tree encompassing seven major areas covering over 200 categories and over 800 tasks, which covers diverse capabilities such as question answering, reasoning, multi-turn dialogue, and text generation, to evaluate LLMs in a comprehensive and in-depth manner. We also design detailed evaluation standards and processes to facilitate consistent, unbiased judgments from human evaluators. A test set of over 3,000 instances is released, spanning different difficulty levels and knowledge domains. Our work provides a standardized methodology to evaluate human alignment in LLMs for both English and Chinese. We also analyze the feasibility of automating parts of evaluation with a strong LLM (GPT-4). Our framework supports a thorough assessment of LLMs as they are integrated into real-world applications. We have made publicly available the task tree, TencentLLMEval dataset, and evaluation methodology which have been demonstrated as effective in assessing the performance of Tencent Hunyuan LLMs. By doing so, we aim to facilitate the benchmarking of advances in the development of safe and human-aligned LLMs.

Cross-domain sequential recommendation (CDSR) aims to enhance recommendation performance in a target domain by leveraging user sequential preferences from a source domain. As most advanced methods employ various neural networks to model item-ID sequences, they often struggle with a sparse data. Recent text-based CDSR approaches utilize rich text information to facilitate knowledge transfer across domains but still encounter three main challenges: (i) how to capture domain-shared semantic relationships; (ii) how to effectively integrate semantic intents and behavior intents; and (iii) how to design sufficient supervision signals for model training. In this paper, we propose a novel text-enhanced CDSR solution, i.e., contrastive text-enhanced Transformer (CTT), which jointly learns user preferences from semantic and behavioral information. Specifically, our CTT is a dual-view structure, including an inter-domain preference view and an intra-domain preference view. In the former, we design a domain-shared cross-attention mechanism to encode semantic correlations across different domains by an attention map, which is then utilized for cross-domain semantic intent learning. In the latter, we extract fine-grained behavior intents from item-ID sequences and coarse-grained semantic intents from item-text sequences, thereby capturing high-quality intra-domain preferences. Our CTT integrates these two types of preferences to predict the next item for each user. Moreover, we propose a semantic-behavior contrastive learning approach to enhance cross-domain preference learning and knowledge transfer. Extensive experiments on three real-world datasets demonstrate that our CTT outperforms the state-of-the-art models by an average of 5.86% on HR@5. The scripts for data preprocessing and conducting experiments, the parameter configurations, and the source codes of our CTT and all baselines are available at https://github.com/XinbeiCai/CTT.