You are expecting too much out of this basic approach. The "simple" similarity search in word2vec (used in https://semantle.com/ https://semantle.com/ if you h

You are expecting too much out of this basic approach. The "simple" similarity search in word2vec (used in https://semantle.com/ https://semantle.com/ if you haven't seen it) is based on _multiple_ embeddings like this one (it's a simple neural network not a simple embedding).