Metaphor Detection Encoder Using GCN and BiLSTM

The patent segments the metaphor detection task into multiple components: word embedding layer, BiLSTM layer for contextual representation, attention mechanism for feature weighting, and classification layer. This segmentation allows each component to specialize in specific aspects of meaning analysis, improving overall detection accuracy and reliability in distinguishing literary from non-literary meanings.

The patent introduces attention mechanisms that add a new dimension to the analysis by dynamically weighting different features and contextual information. The attention layer computes attention scores that highlight relevant features for metaphor detection, enabling the system to focus on discriminative patterns that distinguish literary from non-literary meanings beyond traditional feature spaces.

The patent merges multiple effective components into a unified deep learning architecture: word embeddings are combined with BiLSTM for contextual representation, which is then integrated with attention mechanisms and classification layers. This merging creates a cohesive system where each component reinforces the others, achieving high detection accuracy while managing complexity through unified design rather than separate processing stages.

The BiLSTM layer serves multiple functions simultaneously: it captures contextual information from surrounding words, models sequential dependencies in the text, and provides rich feature representations for the attention mechanism. This multi-functionality reduces the need for separate specialized components, thereby improving detection accuracy without proportionally increasing system complexity.

The patent applies preliminary action by using pre-trained word embeddings (such as GloVe or Word2Vec) that capture semantic relationships before the main processing. This preliminary encoding provides rich initial representations that accelerate subsequent processing while enabling sophisticated metaphor detection, as the model starts with already-computed semantic knowledge rather than learning from scratch.

The BiLSTM layer acts as an intermediary between simple word embeddings and the classification layer. It transforms static word vectors into dynamic contextual representations by processing sequences bidirectionally, capturing both left and right context for each word. This intermediary processing enhances text comprehension accuracy for metaphor detection while maintaining efficient processing through optimized recurrent computations.