inverse operation of word2vec - vec2word

[Julien Petot]

I use word2vec to get the coordinates of words.

I would like to know if it is possible to do the inverse operation of word2vec, i.e. get some words from some new synthetic vector (vec2word).

related post: https://stats.stackexchange.com/questions/422430/inverse-word-embedding-vector-to-word

thanks

[Andrey Kutuzov]

Hi,

The task you are describing is a subset of what is called "definition
modeling". Check this paper for an introduction:
https://ojs.aaai.org/index.php/AAAI/article/view/10996

> --
> You received this message because you are subscribed to the Google
> Groups "Gensim" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to gensim+un...@googlegroups.com
> <mailto:gensim+un...@googlegroups.com>.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/gensim/726c6dbc-df0a-4c5d-9c8a-b8194381dbf8n%40googlegroups.com <https://groups.google.com/d/msgid/gensim/726c6dbc-df0a-4c5d-9c8a-b8194381dbf8n%40googlegroups.com?utm_medium=email&utm_source=footer>.

--
Solve et coagula!
Andrey

[Gordon Mohr]

Gensim's `.most_similar()` method will accept arbitrary vectors as the 'origin' from which it searches for nearby neighbor known words. To avoid confusing its type-detection, such vectors should be passd as part of the explicit `positive` argument's expected list-of-targets (rather than just an ordered argument). For example:

    my_origin = ...  # some properly-dimensioned vector

    similar_words = kv_model.most_similar(positive=[my_origin])

- Gordon