关于本岗位:
用户对影像风格、修图方式和创作意图的选择,构成了个性化体验的基础。本岗位聚焦用户理解与个性化算法的建设,基于多模态内容与用户行为数据,对长期偏好、即时意图和审美兴趣进行建模,并将这些能力应用于搜索推荐、AI创作和Agent个性化场景。从算法设计、工程实现到线上实验,您将参与完整的效果优化链路,使模型改进转化为用户可感知的产品体验。
岗位职责:
· 构建用户偏好与意图理解模型。融合用户行为及图像、文本等多模态数据,构建用户标签、向量表示与兴趣模型,提升对审美偏好、创作需求和实时意图的理解能力。
· 推动用户理解能力落地。将画像与用户表示应用于素材推荐、效果推荐和内容推荐,支持搜索个性化、AIGC创作和用户运营策略,通过离线评估与A/B实验持续验证并优化效果。
· 探索大模型时代的个性化技术。利用大模型进行偏好提取、意图识别和用户信息总结,结合Agent的上下文与长期记忆,提升AI对用户偏好的理解、延续和响应能力。
· 搭建可持续迭代的画像体系。与数据、工程和产品团队协作,完善特征与标签生产、画像更新和质量评估机制,将用户反馈融入算法迭代,在效果、时效和服务成本之间取得平衡。
任职要求:
· 计算机、数学、统计、人工智能等相关专业硕士及以上学历,具有3年及以上算法研发经验。
· 在用户建模、推荐算法、搜索算法或数据挖掘方向有实际项目经验,能够独立完成问题分析、模型设计和效果验证。
· 具备扎实的机器学习基础,熟悉特征工程、分类回归、聚类或向量表示等常用方法,能够根据业务问题选择合适的技术方案。
· 熟练使用Python、Java或Scala中至少一门语言,掌握PyTorch、LightGBM或XGBoost等至少一种框架或工具,具备良好的算法工程实现能力。
· 对数据和业务效果敏感,能够结合评估指标、实验结果与用户反馈定位问题,并与产品、工程团队协作推进落地。
· 对大模型与Agent的用户理解和个性化应用有兴趣,愿意结合实际场景探索和验证新方法。
加分项:
· 具有大规模用户画像、用户向量、兴趣建模或实时意图识别经验。
· 具有搜索推荐的召回、排序、序列建模或冷启动优化经验。
· 具有多模态理解、审美偏好建模,或影像、创作工具、AIGC产品相关经验。
· 具有大模型应用、Agent记忆或个性化交互等项目的落地经验。
· 熟悉Hive、Spark、Flink等数据处理组件。
岗位特点:
· 算法问题具有独特性:从点击偏好深入至审美偏好与创作意图,面临更丰富的用户理解挑战。
· 应用场景明确:算法能力可落地至影像产品、搜索推荐和AI创作流程,并通过线上实验验证价值。
· 技术能力具有延展空间:在用户建模和搜索推荐积累的基础上,可进一步探索多模态、大模型和Agent个性化方向。
如果您擅长搜索推荐或用户建模,并希望进一步探索如何让AI更懂用户,欢迎加入。我们重视候选人在某一方向的扎实积累,以及将算法落实到真实业务中的能力,无需同时具备所有加分项。
About This Role:
Meitu's imaging products, AIGC creation tools, and AI agents serve a diverse global user base. This role sits at the center of our user understanding and personalization efforts. You will model long-term preferences, immediate intent, and aesthetic interests using multimodal content and user behavior, then apply those capabilities to search, recommendation, AI creation, and agent personalization. You will be involved in the full cycle from algorithm design and engineering implementation to online experimentation, turning model improvements into product experiences users can perceive.
Key Responsibilities:
· Build user preference and intent models by fusing behavioral data with image, text, and other multimodal signals, producing user tags, vector representations, and interest models that improve understanding of aesthetic preferences, creative needs, and real-time intent;
· Apply user profiles and representations to asset, effect, and content recommendation, supporting search personalization, AIGC creation, and user operations through offline evaluation and A/B testing;
· Explore LLM-based approaches for preference extraction, intent recognition, and user information summarization, and integrate agent context and long-term memory so AI can better interpret, retain, and respond to user needs;
· Strengthen profile capabilities by collaborating with data, engineering, and product teams to improve feature and tag production, profile freshness, and quality evaluation, incorporating user feedback into iterative algorithm development while balancing effectiveness, timeliness, and service cost.
Requirements:
· Master's degree or above in computer science, mathematics, statistics, AI, or a related field, with 3+ years of algorithm development experience;
· Hands-on project experience in at least one of user modeling, recommendation algorithms, search algorithms, or data mining, with the ability to independently lead problem analysis, model design, and performance validation;
· Solid machine learning foundations, including feature engineering and methods such as classification, regression, clustering, or vector representation, with the ability to select appropriate technical approaches for different business problems;
· Proficiency in at least one of Python, Java, or Scala, and practical experience with at least one framework or tool such as PyTorch, LightGBM, or XGBoost, with the ability to implement algorithms in production-grade systems;
· Sensitivity to data and business outcomes, demonstrated by the ability to identify issues through metrics, experimental results, and user feedback, and to work with product and engineering teams to drive solutions into production;
· Genuine interest in applying LLMs and agents to user understanding and personalization, with a willingness to test and validate new methods in real product scenarios.
Preferred Qualifications:
· Experience building large-scale user profiles, user embeddings, interest models, or real-time intent recognition systems;
· Experience with recall, ranking, sequence modeling, or cold-start optimization in search or recommendation;
· Experience with multimodal understanding, aesthetic preference modeling, or products related to imaging, creative tools, or AIGC;
· Experience applying LLMs, agent memory, or personalized interaction in production systems;
· Familiarity with data processing components such as Hive, Spark, or Flink.
What This Role Offers:
· A distinctive set of algorithm problems that go beyond click preference to aesthetic preference and creative intent;
· Clear application paths into imaging products, search and recommendation, and AI creation, with online experiments to validate real impact;
· Room to extend user modeling and search/recommendation experience into multimodal, LLM, and agent personalization.
If you have a strong foundation in search, recommendation, or user modeling and want to explore how AI can understand users more deeply, we would like to hear from you. We value focused depth and the ability to turn algorithms into real product impact; you do not need to meet every preferred qualification to be considered.