问题驱动

知识图谱：

1. 不区分概念和实例 —> 上下位传递问题；[2018_EMNLP_Differentiating Concepts and Instances for Knowledge Graph Embedding]
2. 知识推理；引入强化学习；虚假样例；虚假路径 [2018_EMNLP_Multi-Hop Knowledge Graph Reasoning with Reward Shaping]

 

 

命名实体识别

1. 自动化标注代价大 [2018_EMNLP_Learning Named Entity Tagger using Domain-Specific Dictionary]

 

文档摘要

1. 单轮抽取 -> 迭代式多轮抽取；选择性读取 [2018_EMNLP_Iterative Document Representation Learning Towards Summarization with Polishing]

 

文本生成

1. seq2seq；同一输入同一输出 -> 同一输入不同输出且有区分性 [2018_EMNLP_Stylistic Chinese Poetry Generation via Unsupervised Style Disentanglement]

 

端到端

1. 引入auto-encoder [2018_EMNLP_Evaluating the Utility of Hand-crafted Features in Sequence Labelling]

迁移学习

1. Domain Generalization；对抗增强迁移方法；[2018_NIPS_Generalizing to Unseen Domains via Adversarial Data Augmentation]

时间序列

1. 结合CNN、RNR、AE。[2018_CIKM_Correlated Time Series Forecasting using Deep Neural Networks: A Summary of Results]

推荐系统

1. 混合专家模型；知识图谱；利用用户序列行为中相邻物品间的关系来解释用户在特定时间点的行为原因，进而基于用户的近期行为对其下一次行为进行预测。[CIKM 2018_Recommendation Through Mixtures of Heterogeneous Item Relationships]
2. 异质信息网络(潜在好友关系、购买关系）；动态采样。[2018_CIKM_Adaptive Implicit Friends Identification over Heterogeneous Network for Social Recommendation] 

转载于:https://www.cnblogs.com/AIlover/p/9981319.html