Learning to Explain Entity Relationships in Knowledge Graphs

Abstract

We study the problem of explaining relationships between pairs of knowledge graph entities with human-readable descriptions. Our method extracts and enriches sentences that refer to an entity pair from a corpus and ranks the sentences according to how well they describe the relationship between the entities. We model this task as a learning to rank problem for sentences and employ a rich set of features. When evaluated on a large set of manually annotated sentences, we find that our method significantly improves over state-of-the-art baseline models.

Dimensions of Interpersonal Relationships: Corpus and Experiments

“…Besides extracting relations per se, there have been efforts within computational linguistics involving interpersonal relationships. Voskarides et al (2015) extract human-readable descriptions of relations in a knowledge graph by ranking sentences that justify the relations. Iyyer et al (2016) propose an unsupervised algorithm to extract relationship trajectories of fictional characters, i.e., how interpersonal relationships evolve over time in fictional stories.…”

A Machine Learning Filter for the Slot Filling Task

“…The part-of-speech tags are generalized into four categories: nouns, verbs, adjectives and other (L1) [28]. If we take Figure 3, for example, the distribution of POS tags for the sentence segment before the query entity ("Nicole/noun Kidman/noun and/other") is the following: nouns: 2/3, verbs: 0/3, adjectives: 0/3, other: 1/3.…”

“…This subset of features captures the location of named entities in the justification relative to the query and the filler for the following types: person (N1), geo-political entity (N2) and organization (N3) [28]. Similarly to feature C4, this set of features is based on the segmentation of the sentence according to the position of the query entity and the slot filler.…”

“…The first segment contains one entity of type person (Nicole Kidman), the second segment does not contain any named entity and the final segment contains one named entity of type geo-political entity (Hawaii). Named-entity features N1 #person left/between/right Number of person left/right or between entities in the sentence [28] N2 #gpe left/between/right Number of Geo-political entities left/right or between entities in the sentence [28] N3 #orgs left/between/right Number of organizations left/right or between entities in the sentence [28] Lexical ( …”

Finding Relevant Relations in Relevant Documents

“…This is different from the task of explaining relationships between entities in a knowledge base [9], since we include also yet unknown facts from documents. It is also different from explaining the relationship between entities and ad-hoc queries [2], since we look at relations between entities in documents.…”