Fig. 3
From: Composition-driven symptom phrase recognition for Chinese medical consultation corpora
Description of the skip-gram model. The model used in Word2Vec to find an optimal representation to predict the surrounding context of a target word. Consider a standard symptom phrases from the symptom dictionary,
(ENG: “supraclavicular lymph nodes were not palpable and enlarged”). The example highlights the window around
(ENG: “lymph node”), organs that produce immune cells for fighting infections. The target word,
(ENG: “lymph node”), is linked to each of its neighboring words and the pairs are fed into the network. The learning process optimizes the probability of predicting the contextual words of
(ENG: “lymph node”)