论文链接 http://link.springer.com/book/10.1007/978-3-319-55753-3?page=2#toc

  1. Learning the Structures of Online Asynchronous Conversations
    挖掘对话中实际的顺序,而不是按时间序列。采用的是豆瓣数据验证(goal是挖掘对话like wechat,但实际上只有豆瓣数据做验证)
    方法: 矩阵, 目标函数, 时间解决的问题: 找到message n 所回复的 message m, TF-DFmodel? 短对话而非normal
    存在的问题: 回复的关系是一条消息只对应一个回复对象(而不是一对多)
    清华大学的学生,报告很有水平,口语很好
  2.  A Question Routing Technique using Deep Neural Network for Communities of Question Answering
    QR-DSSM
    目标: 根据所要解决的问题,rank相关的专家来回答问题。(routing)
    Ranking Model, SVM, LDA
    data: stack overflow问答社区
  3. A General Fine-Grained Truth Discovery Approach for Crowdsourced Data Aggregation
    目标: 通过多个layer矩阵,分配众包
    LDA主题模型, 通过聚类方法得到多个聚类,然后将聚类作为对象建立矩阵关系
    矩阵关系直接是怎么联想的?最后目标是什么?
  4. Category-Level Transfer Learning from Knowledge Base to Microblog Stream for Accurate Event Detection
    without adjusting the threshold, LDA作为对比对象
    问题(目标): transfer KB's category-level info into microblog stream
    1) Extrating Category-level Topics in KB
    2) Transferring category-level info into Microblog Stream
    3) Detecting Events on Category-level ...
    数据: 知识库,维基百科


Data Mining => 挖掘数据中蕴含的模式?

  1. Efficiently Discovering Most-Specific Mixed Patterns From Large Data Trees
    动机:不是简单tree而是混合tree pattern (当前大多数方法都是简单tree)
    mixed pattern 能够融入更多信息和细节(用图展示了)
    input data, morphism, pattern
    方法:
  2. Fast Extended One-Versus-Rest Multi-label SVM Classification Algorithm Based on Approximate Extreme PointsSemi-Supervised Network Embedding
  3. Max-Cosine Matching Based Neural Models for Recognizing Textual Entailment
    讲 Recognizing textual entailment
  4. An Intelligent Field-aware Factorization Machine Mode
    论文有待阅读,启发:英语口语很重要。


Recommendation

  1. Leveraging Kernel Incorporated Matrix Factorization for Smartphone Application Recommendationn
    Kernel function; 基于PMF进行修改; 融入category-info; 梯度下降求解;根据用户安装的APP数据推荐用户可能感兴趣的APP.
  2. Perference integration in context-aware recommendation
    Group Divisionn Tree (减去不需要的信息,有些信息反而会影响最终的结果) Incorporate two-layer preference to provide the recommendatin lists. 这篇文章的做法有点像GTRM。三维的实验结果图看起来不错。
  3. Jointly Modeling Heterogeneous Temporal Properties in Location Recommendation
    融合多个时间粒度(月,星期,天,小时等)的地点推荐。(EM求解)
  4. Location-Aware news recommendation Using Deep Localized Semantic Analysis
    物理距离,主题相似距离;
    改进ELSA模型?(Explicit Localized Semantic Analysis)
    Wikipedia-based topic model?跟距离有什么关系
    提出了三种方法:
    CLSA: Clustering-based localized Sementic Analysis
    ALSA: Autoencoder-based
    DLSA: Deep Localized Semantic Analysis
    Twitter dataset and Wikipedia snapshot to construct topic space.
  5. Review-Based Cross-Domain Recommendation through Joint Tensor
    Domain-unbalanced?
    transfer model? 传统方法在transfer方面的信息局限于用户信息,这篇文章不太一样,不一样在哪?
    transfer user latent factors
    transfer aspect latent factorsoveall rating prediction


留言

登录 请先登陆, 再留言!