Semantic Similarity

This app calculates a measure of semantic similarity (dot product of the word embeddings) between a target word and a list of "positive" attribute word versus a set of "negative" attribute words. Uses word2vec and the gensim package.

A contrast word can be specified, in which case similarities are calculated with the target - contrast vector.

This was used as part of the method in Gladwin (2022). One of the ideas for future research was to use measures like this to select target words that, hypothetically, either (1) have a strong model-based (i.e., common, population-level) association with one versus the other attribute category to get strong within-subject effects or (2) actually have a weak relative association to allow for more individual variability and hence higher reliability.

Input

Please enter the words as a comma-separated list.

Target words:

Target-contrast word:

Positive words:

Negative words:

Example

An example to run a list of emotional words through a "valence filter": You should see the positive emotions get positive similarities and the negative emotions get negative similarities.