项目名称: 考虑用户浏览行为的网络短文本推荐的研究
项目编号: No.61202213
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 计算机科学学科
项目作者: 刘盛华
作者单位: 中国科学院计算技术研究所
项目金额: 24万元
中文摘要: 在社会化网络时代,承载着丰富信息的短文本数据被大量地产生出来,并在互联网络上迅速传播。根据用户浏览行为产生的数据,推荐用户感兴趣的短文本信息及其链接,成为帮助用户及时发现和有效地获取各种网络媒体信息的重要手段。这种以网络短文本数据为主要内容的推荐,给传统的推荐技术及其所依赖的基本假设都带来了新的挑战,并逐渐引起工业界和学术界的广泛关注。本课题的研究针对短文本篇幅短包含的内容极少、流式和时效性短、以及特征的非独立性等挑战,以微博话题的动态推荐和基于用户浏览历史的网页浏览推荐为基本应用场景,将内容特征极稀疏的高维数据建模、流式短文本数据的在线话题聚类、非独立多特征数据的推荐模型等作为课题主要的研究内容,来解决上述问题。本课题的研究旨在帮助解决,利用用户浏览行为的数据进行网络短文本推荐所面临的基本问题,进一步完善推荐技术,推动推荐系统在社会化互联网中更广泛和有效的应用。
中文关键词: 短文本;话题模型;情感分类;信息传播;推荐
英文摘要: In the age of Social Web, short texts are prevalent on the web, such as microblogs, web videos/photos' brief, news brief, etc., which carris massive information and spreads aggresively through the internet. Recommending interesting short texts with their links using the data from user browsing behaviors, plays an important role in helping users to discover and access useful information effeciently. Nevertheless, such a recommendatoin with short texts as its objects, brings new challenges to traditional recommendatoin technologies and their basic hypotheses, which attracts more and more attentions from both academics and industry. Based on the scenarios of topical recommendation on microblog, and browsing recommendation on user's browsing history, the proposal considers those challenges of extremely short content, realtime streaming data, multple dependent feathures, and proposed three key research topics to tackle those challenges. Those research topics include modeling on extremely sparse and high dimentional data, on-line clustering realtime streaming data, and recommendation on the data with multiple dependent features. The purpose of the proposal aims at solving those basic problems and challenges from recommedation short text considering browsing behaviors, which help improve the current recommendation tech
英文关键词: short text;topic model;sentiment classification;information propagation;recommendation