A gated recurrent unit (GRU) processes words sequentially, retaining important context from earlier words so later words can be interpreted. In a classification pipeline, it follows word embeddings and SpatialDropout1D, then passes its learned representation to a Dense layer with sigmoid output. section 4.6.6
Text classification pipeline
Words such as "nausea" and "vomiting" become vectors, pass through dropout and a GRU, and produce a probability between 0 and 1 for a classification task. The surrounding pipeline also compensates for rare high-risk cases with class weights. section 4.6.6
Figure 4.12 The classification pipeline architecture

Use the pipeline architecture figure to trace data from word embeddings through the GRU to the final classifier. section 4.6.6









