Negative sampling
A common method to estimate the conditional probability (Language model) in word2vec and similar methods.
It is actually a biased estimator. See Noise Contrasive Estimation suggested by Gutmann2010noise. But due to this bias, it estimates another reasonable conditional probability that takes into account the frequency of the target word. See Kojaku2021Residual2Vec for more details.
Liu2024re visiting studies dimension regularization, which leverages the idea that node repulsion is similar to making them more orthogonal (dimension regularization).