An effective deep learning based Idrcnn and Bdc-Lstm models for complex word identification and synonym generation

Finding words in a document that the average reader could find challenging to understand is usually the first step in identifying complex words in the text. After you've determined which terms are difficult, you might create synonyms or offer substitutes that would be simpler to read to improve the text's accessibility. As a result, inaccurate synonyms unrelated to the context may be created. In this field of study, combining the recognition of difficult words with the creation of synonyms is novel. We provide a unique deep-learning technique and algorithm to address the problems in the prior research's difficult word identification and synonym-generating tasks. This research is evaluated on the Complex Word Identification dataset. Initially, stop word removal, tokenization, punctuation removal, lowercasing, and stemming are performed in the pre-processing phase. After that, utilize the FastText method to perform word embedding on pre-processed input text. Then, the attributes of morphological and linguistics from embedded text are retrieved to enhance the classification performance. Then, recognize the complex words from the given text utilizing the Improved Deep Residual Convolutional Neural Network (IDRCNN) approach and also employ the Single Candidate Optimization Algorithm (SCOA) to enhance the classification performance by tuning and optimizing the parameters of the technique. Finally, utilize the Bi-Directional Convolutional Long Short Term Memory (BDC-LSTM) for synonyms generation for those identified words. The experimental results demonstrate that the proposed approach performs better than the alternative approaches in terms of entity synonym set construction performance.