Sample Approach

Sample Approach


Jun. 22, 2018
Here are introduced some approaches that can be helpful to enter the Challenge (spoiler-free).

Searching Approach

Identification of the criminal by searching with SPARQL queries

Spot a person who was present at a certain place and time.
This is an approach introduced at JSAI2018@Kagoshima. In order to identify the one who has a motive for murder, characters were listed up with reference to the time when the incident occurred and the abstract time.
You may need, however, to add some other facts and rules to conclude what can be the motive of murder. In the Challenge, you can add facts and rules freely.

Reasoning Approach

Reasoning by additional rules based on a first order predicate logic

His/her Fingerprints were found on the knife. He/she is the criminal?
It is clear who the criminal in this story is. Nevertheless, reasoning rules should be created not only to be able to trace this criminal, but to be applicable to other stories (applicability). You may make good use of a small number of rules for reasoning, or you may reason with fewer steps using a large number of rules.

Reasoning with the description theory based on ontology

Prepare a simple ontology and introduce a class definition into the rules.
In preparation of a set of general rules, try considering the use of ontology. (It is important.)

Machine Learning Approach

Categorization of the criminals using supervised learning

Learn and reason from the characterizing group of the criminals in other cases.
Although only one knowledge graph from one story is published now, you may be able to pick up factors that lead to the identification of the criminal (financial matters, feelings of resentment, etc.) from other movies, dramas or stories, and use them as supervising data to generate a model. And then, in the model, you enter the characteristics of each person in this story to estimate the probability of being the criminal. In this case, you are required to clarify which characteristics was the most effective factor in estimating the criminal (explainability in machine learning).

Clusterization using unsupervised learning

Learn from the characterizing group of the criminals in other cases.
Once you have made a list of characters and characterizing groups from other movies, dramas and stories, it may be an interesting attempt to use a clustering approach entering the characters and their characteristics of this story. The criminal cluster may include the criminal of this case.