Retweet or not?: personalized tweet re-ranking
Retweet or not?: personalized tweet re-ranking
复制标题
DOI：
10.1145/2433396.2433470
复制
发表时间：
2013-02
期刊：
影响因子：
--
通讯作者：
W. Feng;Jianyong Wang
中科院分区：
文献类型：
--
作者：
W. Feng;Jianyong Wang
随着Twitter在世界各地的广泛使用，用户每天都会面临大量的新推文。推文按时间顺序进行排名，而不考虑它们的潜在兴趣。用户必须浏览推文页面才能找到有用的信息。因此，需要更个性化的排序方案来过滤淹没的信息。由于转发历史揭示了用户对推文的个人偏好，因此我们研究了如何学习预测模型来根据推文被转发的概率对推文进行排序。通过这种方式，用户可以在短时间内找到有趣的推文。为了模拟转发行为，我们构建了一个由三种类型的节点组成的图：用户、发布者和推文。为了融合所有信息源，如用户的个人资料、推文质量、交互历史等，节点和边由特征向量表示。所有这些特征向量都被映射到节点权重和边权重。基于该图，我们提出了一种特征感知的因子分解模型来对推文进行重新排序，该模型将线性判别模型和低阶因子分解模型无缝地结合在一起。最后，我们在从Twitter上抓取的真实数据集上进行了广泛的实验。实验结果表明了该模型的有效性。
With Twitter being widely used around the world, users are facing enormous new tweets every day. Tweets are ranked in chronological order regardless of their potential interestedness. Users have to scan through pages of tweets to find useful information. Thus more personalized ranking scheme is needed to filter the overwhelmed information. Since retweet history reveals users' personal preference for tweets, we study how to learn a predictive model to rank the tweets according to their probability of being retweeted. In this way, users can find interesting tweets in a short time. To model the retweet behavior, we build a graph made up of three types of nodes: users, publishers and tweets. To incorporate all sources of information like users' profile, tweet quality, interaction history, etc, nodes and edges are represented by feature vectors. All these feature vectors are mapped to node weights and edge weights. Based on the graph, we propose a feature-aware factorization model to re-rank the tweets, which unifies the linear discriminative model and the low-rank factorization model seamlessly. Finally, we conducted extensive experiments on a real dataset crawled from Twitter. Experimental results show the effectiveness of our model.