Extracting Relations in Texts with Concepts of Neighbours

Abstract

During the last decade, the need for reliable and massive Knowledge Graphs (KG) increased. KGs can be created in several ways: manually with forms or automatically with Information Extraction (IE), a natural language processing task for extracting knowledge from text. Relation Extraction is the part of IE that focuses on identifying relations between named entities in texts, which amounts to find new edges in a KG. Most recent approaches rely on deep learning, achieving state-ofthe-art performances. However, those performances are still too low to fully automatize the construction of reliable KGs, and human interaction remains necessary. This is made difficult by the statistical nature of deep learning methods that makes their predictions hardly interpretable. In this paper, we present a new symbolic and interpretable approach for Relation Extraction in texts. It is based on a modeling of the lexical and syntactic structure of text as a knowledge graph, and it exploits Concepts of Neighbours, a method based on Graph-FCA for computing similarities in knowledge graphs. An evaluation has been performed on a subset of TACRED (a relation extraction benchmark), showing promising results.

A Two-Step Approach for Explainable Relation Extraction

“…Given a sentence (e.g., "Berlin became the capital of Germany in 1990"), two named entities in the sentence (e.g., "Berlin" and "Germany"), and the assumption that there is a relationship between the two entities, the problem is to predict the label of the relationship (e.g., "is the capital of"), and to provide interpretable explanations for the predicted label. The work presented in this section is developed in more details in [1]. [5] is a graph mining method for entity-relation graphs that aims, for a given tuple of entities, to compute which are the most similar tuples of entities.…”

“…Figure 2 shows the modeling of the sentence "The University of Rennes is French". We rely on NLP tools and resources to extract syntactic and semantic information from text 1 .…”

“…Unfortunately, deep learning methods only predict relationships at the instance level, they do not provide information that can be leveraged into general and interpretable extraction rules. In previous work [1] a symbolic approach based on Concepts of Neighbours [5] was proposed to provide explainable predictions. Those explanations have the potential to be translated into extraction rules.…”