Word Embedding Based Paraphrase Generator for Kannada Language

Hits: 127
Research areas:
Year:
2021
Type of Publication:
Article
Keywords:
Paraphrase Generation, Skip Gram Model, Cosine Similarity, Word Vector, Neural Network
Authors:
Ashwini Gadag; Dr. N. K. Cauvery
Journal:
IJAIM
Volume:
9
Number:
6
Pages:
102-109
Month:
May
ISSN:
2320-5121
Abstract:
The process of paraphrase generation is generating paraphrase sentences from given input sentence with more clarity in content of information. Training large corpus by deep learning is emerging technology in the field of Natural Language Processing. In the proposed work, artificial neural network algorithm is used to determine the semantic similarity between two pieces of text. The word vectors were trained by using the traditional Continuous Skip gram model. The two different words have similar word vector, then model will say two words are contextually similar. The neural networks inform us the probability of each word in the dictionary of being nearby word that we are interested. Probabilities of the output relate to how probable it is to find each dictionary word nearby for the input word. Paraphrase generation of rare words is performed using subword units obtained by segmenting words using the byte-pair encoding compression algorithm. The model is trained and tested for the Kannada news articles.
Full text: IJAIM_639_FINAL.pdf

Indexed By