Query rewriting is a core component of web search, yet traditional methods mainly rely on large language model (LLM) prompting, fine-tuning, or personalized rewriting based on user history. These approaches often overlook population-level intent signals in large-scale query logs, leading to misalignment with mainstream search intent. Moreover, although large models achieve high rewriting quality, their computational demands and deployment complexity limit industrial applicability. To address this, we propose Population-Guided Intent-Aware Rewriting (PGIR), which captures dominant population intent via a Semantic Clustering Unit (SCU) and generates intent-aware rewrites through a Rewriting Unit (RU), aligning queries with mainstream search goals without requiring user history.Building upon PGIR, we further introduce PGIR-DPA, a dual-phase adaptation strategy that transfers capabilities from a teacher LLM to lightweight student models, achieving high-quality rewriting while ensuring industrial scalability. Extensive offline and online experiments, including A/B testing on Baidu Search, show substantial improvements in rewrite quality and a 6.28% relative increase in user satisfaction. The framework has been fully deployed in Baidu's production search system, operating stably at scale, validating its industrial feasibility and commercial value.

Warning: This work contains content that may be offensive or upsetting.Eliminating toxicity from large language models (LLMs) is critical to ensure user safety.However, current methods suffer limitations in the analysis and utilization of toxic samples, failing to fully harness their potential.Through comparative analysis of toxic and safe samples, we identified that (i) toxic samples exhibit diversity and (ii) there lies specificity within this diversity.These findings suggest that leveraging these characteristics of toxic samples could enhance the performance of algorithms in LLMs detoxification.Thus, we propose a novel diverse detoxification framework, Di-vDetox, which comprises two innovative components: a Multi-Category-Induced Personalized Sample Generation (MPSG) strategy and a Scaled Contrastive Direct Preference Optimization (SC-DPO) approach.The former is designed to elicit a variety of personalized toxic responses from LLMs, while the latter is constructed to precisely and fully utilize these toxic responses.Experiments on benchmark datasets across different model scales and various detoxification tasks confirm the effectiveness of our architecture.Our codes are available at https: //github.com/zy1998-c/DivDetox.

Abstract With the spread of national sports awareness, more people are fond of outdoor mountaineering. Intelligent mountaineering equipment can provide some safe and convenient services for outdoor mountaineering. However, the positioning accuracy is affected by the ability of remote sensing image recognition in the background. To address the haze interference in remote sensing image positioning, the Retinex algorithm is enhanced using an atmospheric scattering model. This improved Retinex algorithm adopts a multiscale retinal enhancement algorithm with color restoration to enhance images. The Gaussian kernel function plays a filtering role, while the guided filtering is used to improve image texture and details. Test data from background remote sensing images of mountaineering equipment are used to evaluate the algorithm. Results indicate that when the entire test set is used for calculations, the normalized information entropy of I-Retinex, Retinex, generative adversarial networks, and Alex models is 0.92. The median values of normalized mean squared error and mean absolute error are 0.13 and 0.15, respectively, outperforming the contrast defogging model. There is a negative correlation between the normalized peak signal-to-noise ratio and haze noise error in each model. When the entire test set is used, the normalized average gradient of I-Retinex is 0.87, significantly higher than the comparison models. However, the I-Retinex model developed in this study lacks optimal average computation time and memory consumption data. Experimental results demonstrate that the improved haze removal model effectively removes haze from remote sensing images, supporting the remote sensing image-related service functionalities of mountaineering equipment.

The study of transmembrane proteins (TMPs) and their binding activities holds significant importance in the pharmaceutical industry. Due to their physicochemical properties, known binding information regarding TMPs remains comparatively scarce and prevented researchers to dig information from known samples. However, research into general binding structure basis can circumvent this barrier and provide more mechanism insights, which we previously demonstrated its existence and named it as TMPs binding Domain. In this study, we try to discover the TMPs binding domain more precisely. Through atomic-level heterogenous graph convolutions, we significantly improved the classification performance of binding domains. This lays the algorithmic groundwork for utilizing binding domains in the study of TMPs binding activities and further boost the drug target research or new drug development.