3 ms·
Dynamic word embedding for evolving semantic discovery
- kbob 9y agoCan someone explain what a "norm" is in this context? > The learned word vector norms across times grow with word frequency, and can be viewed as a time series for detecting trending concepts with more robustness than word frequency. > > Generally, comparing to frequencies which are more sporadic and noisy, we note that the norm of our embeddings encourages smoothness and normalization while being indicative of the periods when the corresponding words were making news rounds. Thanks.