Application of RXD Algorithm to Word Vector Representation for Keyword Identification
Alex Sumarsono · 2020 10th Annual Computing and Communication Workshop and Conference (CCWC) · 2020
Word embeddings in Natural Language Processing have seen exponential growth in recent years. Various methods have been developed to map words into vector space representations while preserving semantic and syntactic relationships. These methods have been successfully applied to a variety of applications, such as information retrieval and document classification. One such method, based on a co-occurrence matrix, is Global Vectors for Word Representation (GloVe). This paper proposes to apply RXD algorithm, an unsupervised hyperspectral image processing anomaly detection algorithm, to the vector representation of words generated by GloVe. The output will be keywords that identify some of the essential ideas in a corpus. Experimental results show that RXD outperforms the standard cosine similarity method in finding anomalous words that capture important ideas contained in documents represented by multi-dimensional word vectors.