# ConceptNet 5.5: An Open Multilingual Graph of General Knowledge

**Authors:** Robyn Speer, Joshua Chin, Catherine Havasi  
**arXiv:** [1612.03975](https://arxiv.org/abs/1612.03975)  
**Date:** 2016-12-12  
**Summary page:** https://www.summarizepaper.com/en/paper/1612.03975v2/

## Key points

- ConceptNet 5.5 is a linked open data resource that enhances machine learning about language.
- It is designed to be used with modern natural language processing (NLP) techniques like word embeddings.
- ConceptNet connects words and phrases in natural language through labeled edges.
- It gathers knowledge from expert-created resources, crowd-sourcing, and games with a purpose.
- Its primary objective is to represent the general knowledge required for understanding language.
- Combining ConceptNet with word embeddings provides a deeper level of understanding than distributional semantics or narrower resources like WordNet or DBPedia.
- The effectiveness of this approach is validated through state-of-the-art results on intrinsic evaluations of word relatedness.
- This leads to improvements in various applications utilizing word vectors, such as solving SAT-style analogies.

## Summary

The paper titled "ConceptNet 5.5: An Open Multilingual Graph of General Knowledge" introduces a new version of ConceptNet, a linked open data resource that enhances machine learning about language by providing specific knowledge and external information. This updated version is specifically designed to be used with modern natural language processing (NLP) techniques like word embeddings. ConceptNet is a knowledge graph that connects words and phrases in natural language through labeled edges. It gathers its knowledge from various sources, including expert-created resources, crowd-sourcing, and games with a purpose. Its primary objective is to represent the general knowledge required for understanding language, thereby improving the performance of natural language applications by enabling them to better comprehend the meanings conveyed by people's words. The authors demonstrate that when ConceptNet is combined with word embeddings obtained from distributional semantics models such as word2vec, it provides applications with a deeper level of understanding that cannot be achieved solely through distributional semantics or narrower resources like WordNet or DBPedia. The effectiveness of this approach is validated through state-of-the-art results on intrinsic evaluations of word relatedness, which subsequently translate into improvements in various applications utilizing word vectors. One such application highlighted in the paper is solving SAT-style analogies. In summary, this research presents an enhanced version of ConceptNet that leverages external information and specific knowledge to improve machine learning about language. By combining ConceptNet with word embeddings acquired from distributional semantics models, the authors demonstrate significant advancements in understanding word meanings and relatedness, leading to improved performance in various natural language processing tasks.