Reasoning Web. Explainable Artificial Intelligence by Unknown

Author:Unknown
Language: eng
Format: epub
ISBN: 9783030314231
Publisher: Springer International Publishing

3.5 Challenges in Rule Mining

Today, Horn rules can be mined efficiently on large KBs [68]. However, many challenges remain.

Negation. KBs usually do not contain negative information. Therefore, it is difficult to mine rules that have a negated atom in the body or in the head, such as . Newer approaches use a variant of the PCA [55], class information [22], or new types of confidence measures [16].

External Information. Since KBs are both incomplete and lacking negative information, it is tempting to add in data from other sources to guide the rule mining. One can e.g., add in information about cardinalities [56], or embeddings computed on text [32].

Numerical Rules. We can imagine rules that detect numerical correlations (say, between the population of a city and the size of its area), bounds on numerical values (say, on the death year of Ancient Romans), or even complex numerical formulae (say, that the ratio of inhabitants of the capital is larger in city states) [18, 24, 48].

Scaling. While today’s algorithms work well on large KBs, they do less well once we consider rules that do not just contain variables, but also entities. Furthermore, KBs grow larger and larger. Thus, scalability remains a permanent problem. It can be addressed, e.g., by smarter pruning strategies, parallelization, or by precomputing cliques in the graph of the KB.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.