User Friendly Automatic Construction of Background Knowledge: Mode Construction from ER Diagrams

Where to Read

Overview: Modes Made Easy

Annotated entity-relationship diagrams suggesting that the Tenure status of a Professor is related to the grades of students they advise and teach.

Modes are a way of way of expressing knowledge about the relationships between entities. Something similar to them is used in various relational reasoning and probabilistic logic learning frameworks in order to constrain the search space for possible hypotheses.

Perhaps you have social network data involving people, who they are friends with, their smoking status, and whether they developed lung cancer. The modes to describe their relationships would look like this:

useStdLogicVariables: true.
setParam: treeDepth=4.
setParam: nodeSize=2.
setParam: numOfClauses=8.
mode: friends(+Person, -Person).
mode: friends(-Person, +Person).
mode: smokes(+Person).
mode: cancer(+Person).
bridger: friends/2.

Unless you’re an expert in Inductive Logic Programming, it might not be immediately obvious what these mean. The key idea of this paper is that we can encode the same information using Entity Relationship Diagrams, which should be familiar to anyone who has written SQL queries or is familiar with relational databases.

This also allows a user to specify Advice in terms of what variables will (or will not) be useful for learning a relationship, providing a means for domain experts to encode their knowledge about the system.

How to Cite

Alexander L. Hayes, Mayukh Das, Phillip Odom, and Sriraam Natarajan. 2017. User Friendly Automatic Construction of Background Knowledge: Mode Construction from ER Diagrams. In Proceedings of the Knowledge Capture Conference (K-CAP 2017). ACM, New York, NY, USA, Article 30, 8 pages. DOI: https://doi.org/10.1145/3148011.3148027

bibTeX:

@inproceedings{Hayes2017UserFriendly,
   author = {Hayes, Alexander L. and Das, Mayukh and Odom, Phillip and Natarajan, Sriraam},
   title = {User Friendly Automatic Construction of Background Knowledge: Mode Construction from ER Diagrams},
   booktitle = {Proceedings of the Knowledge Capture Conference},
   series = {K-CAP 2017},
   year = {2017},
   isbn = {978-1-4503-5553-7},
   location = {Austin, TX, USA},
   pages = {30:1--30:8},
   articleno = {30},
   numpages = {8},
   url = {http://doi.acm.org/10.1145/3148011.3148027},
   doi = {10.1145/3148011.3148027},
   acmid = {3148027},
   publisher = {ACM},
   address = {New York, NY, USA},
}