Knowledge Graph and Semantic Computing. Knowledge Computing and Language Understanding by Unknown

Author:Unknown
Language: eng
Format: epub
ISBN: 9789811331466
Publisher: Springer Singapore

(5)

where denotes the log-uniform sampler that samples the number of labels from entity set . is defined analogously. It is worth noting that, this sampler needs the labels in a lexicon sorted in descending order of frequency, thus we should also separately calculate the frequencies of entities and relations.

3.4 Enhancing Entity Prediction with Relation Prediction

Due to the input is length-3 triples, the model only minimizes two sub-losses for each triple. Given a triple (s, r, o), the model learns to predict r based on s, and to predict o based on . We propose a method that can leverage relation prediction for enhancing entity prediction. In Sect. 5.1, the experimental analysis proves that learning to predict relations is helpful for entity prediction.

Reversing relations is a commonly-used method to enable KG completion models to predict head and tail entities in an integrated fashion [10, 14]. Specifically, for each triple (s, r, o) in the training set, a reverse triple is constructed and added into the training set. Thus, a model can predict tail entities with input (s, r, ?), and predict head entities with .

Previous models for KG completion need s, r to predict o. However, the ability of predicting relations enables our model to evaluate the probability distribution of reverse relations for each entity. For example, given an entity , if the probability of is very close to zero, then we can speculate that does probably not have the relation . In other words, is not an appropriate prediction for (s, r, ?). We formulate this by the following equation:

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.