A method for constructing word sense embeddings based on word sense induction.

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Machine Learning

Background:

• Polysemy, the existence of multiple meanings for a single word, presents a significant challenge in natural language understanding.
• Accurate disambiguation of word senses is crucial for various NLP tasks, including information retrieval and machine translation.

Purpose of the Study:

• To propose a novel method for encoding multiple senses of polysemous words into a single vector representation.
• To enhance word sense induction and the accuracy of word sense embeddings.
• To address the inherent ambiguity of natural language.

Main Methods:

• Utilizing a two-layer bidirectional long short-term memory neural network and a self-attention mechanism to extract contextual word information.
• Applying an improved K-means clustering algorithm, optimized with density peaks clustering and cosine similarity, for word sense induction.
• Generating distinct word sense embedded representations for polysemous words.

Main Results:

• The proposed method demonstrated superior performance in word sense induction compared to traditional metrics like Euclidean distance, Pearson correlation, and KL-divergence.
• The generated word sense embeddings were more accurate than those produced by mean shift, DBSCAN, spectral clustering, and agglomerative clustering algorithms.
• Experimental results validate the effectiveness of the proposed approach in handling polysemy.

Conclusions:

• The developed method successfully encodes multiple word senses into a single vector, offering a robust solution for polysemy.
• This approach significantly improves word sense induction and embedding accuracy, advancing the field of natural language processing.
• The findings provide a valuable contribution to the computational linguistic community for better semantic understanding.