---

[TOC]

Knowledge Outline

• Type: Unstructured、Structured
• Methods: Theory 、Applying

Leveraging Knowledge Bases in LSTMs for Improving Machine Reading, 2017 ACL

Motivation

How to add knowledge base to LSTM
Adaptive attention way

Method

Knowledge-aware Bidirectional LSTMS

Embedding Knowledge Base Concepts

• WordNet and NELL
• (e1, r, e2)
  

Experiments

Entity Extraction

• Data: ACE2005、OntoNotes 5.0
• Results:
  
  

Event Extraction

• Data: ACE2005
• Results:
  

World Knowledge for Reading Comprehension: Rare Entity Prediction with Hierarchical LSTMs Using External Descriptions, 2017 EMNLP

Knowledge Form (Data)

• Wikilinks Dataset -> Wikilinks Rare Entity Prediction
  

Model

• Double Encoder
  

• Hierarchical Double Encoder
  

Results

A Knowledge-Grounded Neural Conversation Model, 2017 arXive

Knowledge Form

• (Named Entities, Free-form Text)
• Twitter, Foursquare

Model Architecture

Yuanfudao at SemEval-2018 Task 11: Three-way Attention and Relational Knowledge for Commonsense Machine Comprehension

Knowledge

• ConcepNet

Model Architecture

Input Layer

• Glove Embeddings, 300d
• Part-of-speech and named-entity embeddings, 12d, 8d
• Relation embeddings, 10d
• Handcrafted Features, 2d

Attention Layer

Output Layer

Experiments

Thinking

• Knowledge Embedding

Feedback&Advice

• Weibo：@伟康青年，@github
• mail：wavejkd@pku.edu.cn