Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
102 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

A Novel Embedding Model for Knowledge Base Completion Based on Convolutional Neural Network (1712.02121v2)

Published 6 Dec 2017 in cs.CL

Abstract: In this paper, we propose a novel embedding model, named ConvKB, for knowledge base completion. Our model ConvKB advances state-of-the-art models by employing a convolutional neural network, so that it can capture global relationships and transitional characteristics between entities and relations in knowledge bases. In ConvKB, each triple (head entity, relation, tail entity) is represented as a 3-column matrix where each column vector represents a triple element. This 3-column matrix is then fed to a convolution layer where multiple filters are operated on the matrix to generate different feature maps. These feature maps are then concatenated into a single feature vector representing the input triple. The feature vector is multiplied with a weight vector via a dot product to return a score. This score is then used to predict whether the triple is valid or not. Experiments show that ConvKB achieves better link prediction performance than previous state-of-the-art embedding models on two benchmark datasets WN18RR and FB15k-237.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (4)
  1. Dai Quoc Nguyen (26 papers)
  2. Tu Dinh Nguyen (19 papers)
  3. Dat Quoc Nguyen (55 papers)
  4. Dinh Phung (147 papers)
Citations (654)

Summary

A Novel Embedding Model for Knowledge Base Completion Based on Convolutional Neural Network

The paper introduces ConvKB, a novel embedding model leveraging convolutional neural networks (CNNs) for knowledge base (KB) completion. ConvKB enhances the capabilities of existing state-of-the-art (SOTA) models by capturing global relationships and transitional characteristics between entities and relations within knowledge bases through a CNN framework.

Overview

Knowledge bases like YAGO and Freebase consist of triples (head entity, relation, tail entity), which are valuable for applications in semantic searching, question answering, and machine reading. However, these KBs are often incomplete, necessitating research on knowledge base completion to predict missing triples.

Traditional embedding models like TransE focus on vector representations of entities and relations to predict the validity of triples by checking their transitional relation properties. ConvKB builds upon this notion by introducing a convolutional layer that processes triples structured as 3-column matrices, thereby exploring global relationships at a more abstract level.

Methodology

ConvKB represents each triple as a matrix. This matrix undergoes convolution operations using filters, each operating over the 3-column format to create feature maps. These maps are consolidated into a single feature vector, which is then multiplied with a weight vector to generate a score, indicating the plausibility of the triple. The novelty lies in the model's ability to effectively utilize the transitional properties of relationships along with the convolutional capture of global relationships.

Results

ConvKB shows superior performance over existing models when evaluated on benchmark datasets WN18RR and FB15k-237. Particularly, ConvKB achieves the highest mean reciprocal rank (MRR) and Hits@10 scores on FB15k-237, highlighting its efficacy in practical link prediction scenarios.

Implications and Future Work

The implications of this research are twofold—both practical and theoretical. Practically, ConvKB can improve the accuracy of KB completion that can directly benefit applications relying on complete and reliable knowledge bases. Theoretically, the integration of CNNs to capture complex entity relationships can inspire future research to explore convolutional frameworks in other domains where data can be abstracted as relational tuples or triples.

Speculatively, future developments might expand ConvKB's architecture to incorporate additional context, such as temporal dynamics or multi-hop relational paths, enhancing its application across more sophisticated and dynamic KB scenarios.

Given the demonstrated results and methodology improvements over baseline models like TransE, ConvKB provides a promising direction for advancing the state-of-the-art in knowledge base completion using deep learning methodologies.