Knowledge Bases

Block seminar, 7 ECTS credits, winter semester 2017–18

Basic Information

  • Type: Block seminar
  • Lecturer: Simon RazniewskiParamita Mirza
  • Credits: 7 ECTS credits
  • Registration: Course is full, unfortunately no further registrations can be accepted
  • Dates: November 16 & 23 (introductory meetings) and to be decided (block seminar).

The seminar is a block seminar and will take place on two (consecutive) days at the end of February or beginning of March--the exact days to be agreed with the participants. There will also be two meetings at the beginning of the semester, for which participation is mandatory.


A detailed list is below

Representation, collection and extraction of general knowledge in knowledge bases (KBs) is at the core of many AI applications. In this seminar, we cover a range of topics around KBs, in particular factual KBs (e.g., Wikidata, YAGO, DBPedia), common-sense KBs (e.g., science knowledge, howto-knowledge, script knowledge) and non-textual KBs (e.g., ImageNet). We explore how these KBs are constructed and how they are used in various applications such as question answering (QA), story/script prediction and biography generation. We also explore learning new facts over KBs.


  • October 4, 2017 -- Registration deadline
  • November 16, 2017 -- Kick-off meeting (participation is mandatory)
    • Time: 11:30 am - 13:00 am
    • Place: Room 23, MPI-Inf building (E 1.4, ground level)
      • Explanation of the structure and organization of the seminar
      • Brief introduction to knowledge bases
      • Presentation of the topics
  • November 23, 2017 -- "How to prepare and present a seminar talk" (participation is mandatory)
    • Time: 11:30 am - 13:00 am
    • Place: Room 23, MPI-Inf building (E 1.4, ground level)
      • As this is a block seminar, it is particularly crucial that the students' presentations are of high quality. This lecture aims at preparing the participants in such a way that their slides and presentations will be of high quality.
  • November 23, 2017 -- students send a ranked list of their top 3 topics via email, and are notified of their assigned topic by the end of the week.
  • December 13, 2017 -- students send a suggestion of the outline of their seminar paper, including an itemization of the planned content for each section.
  • January 31, 2018 -- students submit their final seminar paper.
  • Two weeks before the first day of the block seminar -- students send preliminary slides
  • Two days before the first day of the block seminar -- students send their final slides which they will use in the block seminar
  • Block seminar, day 1: TBD
  • Block seminar, day 2: TBD

Rules and Grading

  • Participation in the kick-off meeting, the "How to prepare and present a seminar talk" lecture, and both days of the block seminar is mandatory.
  • Students will be assigned a particular topic and have to submit a seminar paper (template will be provided) and give a presentation (20 minutes + 10 minutes for discussion) over the topic.
  • Grading will be based on:
    • the report
    • the presentation
    • knowledge on the subject (as evidenced in the discussion after the presentation)
    • activity in the discussions
    • ability to stick to deadlines
  • Attention: According to the study regulations, you are only allowed to withdraw from the seminar within three weeks after the kick-off meeting, i.e., until December 7. Later withdrawal counts as "failed".


Area I: Knowledge Bases

  • 1: Collaborative KBs: Wikidata -- Simon
    • Wikidata: a free collaborative knowledge base (Vrandečić & Krötzsch, CACM 2014) [pdf]
    • Introducing Wikidata to the linked data web (Erxleben et al., ISWC 2014) [pdf]
  • 2: Structured information extraction: DBpedia and YAGO -- Simon
    • DBpedia – A large-scale, multilingual knowledge base extracted from Wikipedia (Lehmann, et al., Semantic Web 2015) [pdf]
    • YAGO2: A spatially and temporally enhanced knowledge base from Wikipedia (Hoffart et al., Artificial Intelligence 2013) [pdf]
  • 3: A General common-sense KB: ConceptNet -- Paramita
    • Representing general relational knowledge in ConceptNet 5 (Speer & Harvasi, LREC 2012) [pdf]
  • 4: Domain-specific activity KBs -- Paramita
    • HowTo KB -- Distilling task knowledge from How-To communities (Chu et al., WWW 2017) [pdf]
    • A hierarchical bayesian model for unsupervised induction of script knowledge (Frermann et al., EACL 2014) [pdf]
  • 5: The Allen AI science challenge & building a science KB: Aristo -- Paramita
    • Domain-targeted, high precision knowledge extraction (Mishra et al., TACL 2017) [pdf]
    • Moving beyond the Turing test with the Allen AI Science Challenge (Schoenick et al., CACM 2016) [pdf]
  • 6: A WordNet-based image KB: ImageNet -- Paramita
    • Imagenet: A large-scale hierarchical image database (Deng et al., CVPR 2009) [pdf]
    • WordNet: A lexical database for English (Miller, CACM 1995) [pdf]

Area II: Learning over Knowledge Bases

  • 7: KB association rule mining -- Simon
    • Fast rule mining in ontological knowledge bases with AMIE+ (Galárraga et al., VLDB 2015) [pdf]
  • 8: Inferring new facts with vector space embeddings -- Simon
    • Learning entity and relation embeddings for knowledge graph completion (TransR) (Lin et al., AAAI 2015) [pdf]
    • Translating embeddings for modeling multi-relational data (TransE) (Bordes et al., NIPS 2013) [pdf]

Area III: Using Knowledge Bases

  • 9: KB question answering -- Simon
    • Automated template generation for question answering over knowledge graphs (Abujabal et al., WWW 2017) [pdf]
    • Robust question answering over the web of linked data (Yahya et al., CIKM 2013) [pdf]
  • 10: Hybrid question answering using KBs and text -- Paramita
    • Question answering on Freebase via relation extraction and textual evidence (Xu et al., ACL 2016) [pdf]
    • Open question answering over curated and extracted knowledge bases (Fader et al., KDD 2014) [pdf]
  • 11: Non-factual QA in the science domain (continuation of Topic 5) -- Simon
    • Answering complex questions using open information extraction (Khot et al., ACL 2017) [pdf]
  • 12: Coreference resolution  -- Paramita
    • Coreference resolution with world knowledge (Rahman & Ng, ACL 2011) [pdf]
  • 13: Referent prediction -- Paramita
    • Modeling semantic expectation: Using script knowledge for referent prediction (Modi et al., TACL 2017) [pdf]
  • 14: Biography generation -- Paramita
    • Learning to generate one-sentence biographies from Wikidata (Chisholm et al., EACL 2017) [pdf]
  • 15: Fact ranking -- Simon
    • The unusual suspects: Deep learning based mining of interesting entity trivia from knowledge graphs (Fatma et al., AAAI 2017) [pdf]

Area IV: Bring your own topic

  • 16: Proposals for other topics are welcome -- Paramita & Simon