The different versions of SenticNet are accessible in RDF/XML format: SenticNet 1 simply associated polarity scores with almost 6,000 ConceptNet concepts; in addition to polarity, SenticNet 2 also assigned semantics and sentics to commonsense concepts and extended the breadth of the knowledge base to about 13,000 entries; SenticNet 3 broadened the spectrum of the semantic network to 30,000 concepts; SenticNet 4 introduced the concept of semantic primitives to further extended the knowledge base to 50,000 entries; finally, SenticNet 5 reaches 100,000 commonsense concepts by employing recurrent neural networks to infer primitives by lexical substitution. The latest version of the knowledge base can also be downloaded in OWL format as an ontology and selectively accessed online through an API, available in 40 different languages. Each language-dependent version of SenticNet is also downloadable in RDF/XML format as BabelSenticNet.


Three vector spaces can also be downloaded in CSV format, namely: IsaCore, a strongly-connected core of hyponym-hypernym common knowledge; AmazonWE, a word2vec model providing 300-dimensional sentiment embeddings generated from Amazon product reviews; and AffectiveSpace, the 100-dimensional vector space representation of AffectNet (a matrix of affective commonsense knowledge, also available for download) upon which SenticNet is built.


We also make some code available on our GitHub account. However, not all the code we develop is freely available: if you are interested in our most advanced tools, please visit our corporate website. In case you need to reproduce some of our results for publication purposes, though, feel free to send us your data and we will do the classification for you.

SenticNet 5
SenticNet 4
SenticNet 3
SenticNet 2
SenticNet 1
Chinese SenticNet
Singlish SenticNet
Amazon embeddings