标签推荐系统中基于深度学习和多目标优化的推荐算法研究

The project, aiming at solving the key problems in tag-aware recommender systems, such as the difficulty of extracting tag information and the difficulty of satisfying users’ multiple requirements, seeks for a more accurate and effective method of tag information extraction, and explores a new way to provide personalized recommendation services that meet different needs of users at the same time. The main research content includes: (1) According to the characteristics of tags, a tag-based deep learning model is constructed to fully mine the potential useful information in tags and obtain the effective abstract feature representation of users and items; (2) Based on the conflicting relationship between accuracy and diversity of recommendation results, a multi-objective recommendation model based on tags is constructed, and an efficient evolutionary multi-objective optimization algorithm is designed to solve the modeled problem, so as to improve user satisfaction; (3) Combining the advantages of multi-objective optimization and deep learning, a tag-based deep multi-objective optimization recommendation algorithm is proposed to improve the accuracy and diversity of recommendation results simultaneously. The project effectively improves the quality of personalized recommendation service by combining advanced machine learning techniques and classical optimization methods, and provides a new idea for the recommendation methods in tag-aware systems.

本项目旨在针对标签推荐系统中存在的标签信息难以提取、用户多种需求难以满足等关键问题，寻求更准确和有效的标签信息提取方法，探索能够提供同时满足用户不同需求的个性化推荐服务的新方法。本项目的研究内容主要包括：（1）针对标签数据的特性，构建基于标签的深度学习模型，充分挖掘标签数据中潜在的有用信息，获取用户和物品的有效抽象特征表示；（2）基于推荐结果的准确度和多样性之间存在的相互冲突关系，构建基于标签的多目标推荐模型，设计高效的进化多目标优化算法进行求解，提高用户的满意度；（3）结合多目标优化和深度学习的优点，提出基于标签的深度多目标优化推荐算法，同时提高推荐结果的准确度和多样性。本项目通过结合先进机器学习技术和经典优化方法，有效提高了个性化推荐服务的质量，为标签系统中的推荐方法提供了新的思路。

本项目旨在针对标签推荐系统中存在的标签信息难以提取、用户多种需求难以满足等关键.问题，寻求更准确和有效的标签信息提取方法，探索能够提供同时满足用户不同需求的个性化推荐服务的新方法。本项目的研究内容主要包括：（1）针对标签数据的特性，构建基于标签的深度学习模型，引入了注意力机制，从而获取用户基于标签的多样的潜在兴趣特征和物品多方面的特征表达，因而能够获得更为有效的潜在特征表示，进而提供更为准确的推荐结果；（2）为了进一步提升用户满意度，考虑用户多样的需求，构建了提高推荐多样性的深度学习模型，融合了多种多样性提升策略，能够显著的提升推荐多样性，同时保持了较高水平的准确度；（3）基于推荐结果的准确度和多样性之间存在的相互冲突关系，构建基于标签的多目标深度推荐模型，结合多目标优化和深度学习的优点，同时优化推荐结果的准确度和多样性，能够为每个用户提供多个不同多样性程度的推荐列表。本项目通过结合先进机器学习技术和经典优化方法，有效提高了个性化推荐服务的质量，为标签系统中的推荐方法提供了新的思路。