Water an article PAKDD2018: topic-specific Retweet Count ranking for WEIBO__PAKDD

Look at the topic and know what to do: topic-specific Retweet Count ranking for Weibo

Summary:

In this paper, we study \emph{topic-specific} retweet the Count ranking problem in Weibo. Two challenges make this task nontrivial. Firstly, traditional methods cannot derive effective for tweets, feature in because topic-specific, tweets setting Have too many shared contents to distinguish them. We propose a lstm-embedded autoencoder to generate tweets features with the insight this any different prefixes of tweet Te The XT is a possible distinctive feature. Secondly, it is critical to fully catch the meaning of topic in topic-specific setting, but Weibo can provide little infor Mation about topic. We leverage real-time news information from Toutiao to enrich the meaning of topic, as more than 85\% topics are News. We evaluate the proposed components based on ablation methods, and compare the overall solution with a recently-proposed t Ensor factorization model. Extensive experiments on real Weibo data show the effectiveness and flexibility to our methods.

It can be seen that this paper is mainly shared in the extraction of topic, tweets, user characteristics of the method. One of the user characteristics of natural existence, do not need to do more processing; On the topic features, because the microblogging itself provides less topic information, this article from today's headlines such a news site to extract the relevant topic information (because research has shown that 85% of the information on Weibo is news, And the attributes of today's headlines are closer), and then using Daes to extract topic features; On the tweet feature, the main problem is that tweets under the same topic are basically the same (including a lot of forwarding, a few comments, short text, etc.), In this paper, the difference between the lstm-embedded Autoencoder and the Autoencoder in machine translation is mainly due to the focus on feature extraction (encoder output) rather than the mapping of two languages (decoder output):

And the entire article uses the sorting method, the Word embedding method all is ready-made, and does not share too much.

Summing up this article sharing there are three points: first, do is topic-specific ranking work, this before very few people do; second, the extraction of tweets, topic methods, although all very intuitive, but can use more scenes; third, the proposed method is not bad.

Find an introductory article for PAKDD2017:

http://data-mining.philippe-fournier-viger.com/pakdd-2017-conference-brief-report/

2) The number of accepted long and short papers in PAKDD forthe last six years is presented below.

5) The acceptance rate of long and short papers in PAKDD during the last six years