How to mismatch words in Conceptnet

Question

Conceptnet contains two main types of nodes, words (e.g. / c / en / cat) and feelings (e.g. / c / en / cat / n / domestic_cat). Unfortunately, the vast majority of edges use dictionary nodes. This makes the conclusion difficult, because I cannot be sure what meaning the word "word for word" refers to.

For example, Conceptnet contains 9 senses that use the word "cat", most of which are correct names (/ c / en / cat / n / musical, / c / en / cat / n / magazine, / c / en / cat / n / a_spiteful_woman_gossip, etc.). If the edge says "/ c / en / cat / r / HasA / c / en / tail", I know, using my own experience, which probably applies to / c / en / cat / n / domestic _cat and no other feelings . If I see an edge that says "/ c / en / cat / r / IsA / c / en / fun_to_watch", I know that it probably refers to / c / en / cat / n / musical, but it also may be referring to / c / en / cat / n / domestic _cat.

How do I automate this process? How to translate edges that use only word nodes so that they use semantic nodes?

Answer 1

ConceptNet does not seem to contain any meaning for the meaning of the word, see their FAQ :

ConceptNet does only the minimum minimum to distinguish words far away - on the built-in ConceptNet 5.5 graphics, the meaning of words is only differing in their part of speech (like semantic 2vec). WordNet has a large number of feelings for each word, although some of them are difficult to distinguish in practice.