This page has been moved to www.ruizhang.info

The Geographic Knowledge Base Project

Brief Description

Knowledge bases in the form of knowledge graphs have wide applications in search engines (such as Google¨s Knowledge Graph), virtual assistants and question answering systems (such as Siri), recommender systems, and natural language generation.

　

This project aims at building, Geographic Knowledge Base (GKB), a knowledge base in a specific domain, geography, with much larger coverage of geographic entities and much richer information than existing generic knowledge bases have. It is a foundational technology for the next generation navigation systems as shown by the example below.

　

　

When you are at the above street location, compare the following two navigation systems:

A GKB is essential for providing the instructions like "turn right at the 3 story red and yellow building" in the next generation navigation systems.

　

In a GKB, we are interested in microscopic geographic entities, which are defined as those geographic entities with the finest level of address such as an apartment building, a house, a stadium, a bridge, a park, etc. A microscopic geographic entity corresponds to a place in Google Maps. The rich information we aim to get include visual features (color, height, materials, architecture style, etc), textual description (functionality, history, etc), and conceptual information such as the importance of a geographic entity.

　

Research Challenges

  • Data sources
  • Entity resolution, duplicates, noises
  • How to increase the number of entities [AAAI'19, ACL'19]
  • How to get high quality information: visual, height [WWW'19], importance
  • What kind of information and relationships
  • How to generate a sentence to describe an entity in a natural way [ACL'18, AAAI'20]
  • Dialogue systems integrating knowledge bases [EMNLP'20]

Datasets

No Task Details Data for download
1 Knowledge Base Alignment README.txt 1. Training data:
      a. Knowledge Base 1 (KB1)
      b. Knowledge Base 2 (KB2)
2. Testing data:
      Entity mapping between KB1 and KB2
2 Relation Extraction for Knowledge Base Enrichment README.txt 1. Train data
2. Validation data
3. Test data
4. Stress test data
3 Sentence Generation from RDF Triples README.txt 1. Train data
2. Validation data
3. Test data
4. Reference

　

Publications

  1. Shiquan Yang, Rui Zhang, Sarah Erfani, GraphDialog: Integrating Graph Knowledge into End-to-End Task-Oriented Dialogue Systems, The 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) 2020.
    　
  2. Bayu Distiawan Trisedya, Jianzhong Qi, Rui Zhang, Sentence Generation for Entity Description with Content-plan Attention. Proceedings of the 33th AAAI Conference on Artificial Intelligence (AAAI 2020), 2020.
    　
  3. Bayu Distiawan Trisedya, Gerhard Weikum, Jianzhong Qi, Rui Zhang. Neural Relation Extraction for Knowledge Base Enrichment. 56th Annual Meeting of the Association for Computational Linguistics (ACL) 2019. [Code and Data]
    　
  4. Yunxiang Zhao, Jianzhong Qi, and Rui Zhang. CBHE: Corner-based Building Height Estimation for Complex Street Scene Images. The Web Conference (WWW), 2019.
    　
  5. Bayu Distiawan Trsedya, Jianzhong Qi, Rui Zhang. Entity Alignment between Knowledge Graphs Using Attribute Embeddings, Proceedings of the 33th AAAI Conference on Artificial Intelligence (AAAI) 2019). [Code] [Data]
    　
  6. Bayu Distiawan Trisedya, Jianzhong Qi, Rui Zhang, Wei Wang. GTR-LSTM: A Triple Encoder for Sentence Generation from RDF Data.  56th Annual Meeting of the Association for Computational Linguistics (ACL) 2018. [Code] [Data]

　

Team

  • Professor Rui Zhang
  • Dr Jianzhong Qi
  • Research Fellow: Bayu Distiawan Trisedya
  • PhD Candidates: Yunxiang Zhao, Shiquan Yang

http://115.146.90.170/

Sponsor