大数据时代新闻推荐模型与方法研究

项目名称： 大数据时代新闻推荐模型与方法研究

项目编号： No.61303081

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 洪文兴

作者单位： 厦门大学

项目金额： 23万元

中文摘要： 大数据技术是当前研究与行业应用的热门问题；新闻信息是大数据时代的一个典型案例；个性化推荐用于解决新闻信息过载和信息迷航问题。结合数据量大、数据类型繁多、时效性严、准确度高的大数据特征，课题提出新的推荐模型和方法，并开发完整的新闻推荐平台，完成"理论、技术、算法和平台建设"四个方面的工作。主要创新点包括：（1）基于多源异构新闻数据的用户兴趣建模。将用户个人信息、阅读历史、社交网络信息、用户反馈等以超图的形式表示，开展基于超图的用户建模研究。在此基础上，依据用户的阅读兴趣，对新闻进行排序和选择。（2）定义子模函数，将新闻推荐转化为预算型最大覆盖问题，确切反应用户阅读兴趣的衰减，结合新闻时效性和地域性，帮助用户选择新闻。（3）从时间、空间、主题和命名实体多个维度展示新闻推荐结果。应用数据可视化技术，对新闻立方体各个维度进行钻取、上卷和切片，以全面展示推荐结果。

中文关键词： 用户建模；新闻建模；个性化；隐变量；

英文摘要： Big Data, a very popular concept in our information era, is attracting increasing attention in both academia and industry. With the challenges of the huge data "Volume", the complex relation-wise "Variety" and the fast updating "Velocity", it is imperative to provide effective and efficient data analysis methods for Big Data processing. News reading domain, as a typical case of Big Data, suffers from the information overload problem, which is often alleviated by personalized recommendation, or called news recommendation. Besides the aforementioned three challenges, news recommendation also requires the information "Veracity", i.e., news users need to read useful or interesting news articles from their own perspectives. To address these challenges, in this project, we propose novel news recommendation models and approaches and implement personalized news recommendation platform. Our goal is to achieve the seamless integration of "theory", "techniques", "algorithms" and "platform". The contribution of this project is as follows: (1). User profiling based on multiple information sources of news data: We propose to use hypergraph to capture the complex relations among news users' personal information, users' reading histories, user's social network information, users' reading feedback, etc. Such a hypergraph provid

英文关键词： personalized recommendation；user modeling；news modeling；latent variable model；