The paper introduces ConceptNet 5.5, an open multilingual graph of general knowledge designed to enhance machine learning about language. ConceptNet connects words and phrases with labeled edges, drawing knowledge from various sources including expert-created resources, crowd-sourcing, and purpose-built games. It aims to improve natural language processing (NLP) techniques, particularly word embeddings, by providing a more comprehensive understanding of word meanings. The paper details the structure and features of ConceptNet 5.5, including its knowledge sources, relations, and term representation. It also describes how ConceptNet can be used as a semantic space and how it integrates with word embeddings to create a hybrid system, ConceptNet Numberbatch. The system is evaluated on word relatedness tasks and downstream applications such as solving SAT-style analogies and choosing sensible endings to stories, demonstrating superior performance compared to other word embedding systems. The results show that ConceptNet Numberbatch outperforms other systems in word relatedness evaluations and performs well on analogies and story cloze tests, highlighting the benefits of combining relational knowledge with distributional semantics.The paper introduces ConceptNet 5.5, an open multilingual graph of general knowledge designed to enhance machine learning about language. ConceptNet connects words and phrases with labeled edges, drawing knowledge from various sources including expert-created resources, crowd-sourcing, and purpose-built games. It aims to improve natural language processing (NLP) techniques, particularly word embeddings, by providing a more comprehensive understanding of word meanings. The paper details the structure and features of ConceptNet 5.5, including its knowledge sources, relations, and term representation. It also describes how ConceptNet can be used as a semantic space and how it integrates with word embeddings to create a hybrid system, ConceptNet Numberbatch. The system is evaluated on word relatedness tasks and downstream applications such as solving SAT-style analogies and choosing sensible endings to stories, demonstrating superior performance compared to other word embedding systems. The results show that ConceptNet Numberbatch outperforms other systems in word relatedness evaluations and performs well on analogies and story cloze tests, highlighting the benefits of combining relational knowledge with distributional semantics.