Please use this identifier to cite or link to this item:
http://hdl.handle.net/20.500.11889/8487
Title: | SemVec: semantic features word vectors based deep learning for improved text classification | Authors: | Odeh, Firas Taweel, Ade |
Keywords: | Convolution Neural Networks,;Artificial intelligence;Text classification;Deep learning (Machine learning);Neural networks (Computer science);Features engineering;Natural language processing (Computer science);Machine learning;Word embeddings | Issue Date: | 2018 | Publisher: | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Abstract: | Semantic word representation is a core building block in many deep learning systems. Most word representation techniques are based on words angle/distance, word analogies and statistical information. However, popular models ignore word morphology by representing each word with a distinct vector. This limits their ability to represent rare words in languages with large vocabulary. This paper proposes a dynamic model, named SemVec, for representing words as a vector of both domain and semantic features. Based on the problem domain, semantic features can be added or removed to generate an enriched word representation with domain knowledge. The proposed method is evaluated on adverse drug events (ADR) tweets/text classification. Results show that SemVec improves the precision of ADR detection by 15.28% over other state-of the-art deep learning methods with a comparable recall score. | URI: | http://hdl.handle.net/20.500.11889/8487 | DOI: | 10.1007/978-3-030-04070-3_35 |
Appears in Collections: | Fulltext Publications |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
SemVec Semantic features word vectors based deep learning for improved text classification.pdf | 338.79 kB | Adobe PDF | View/Open |
Page view(s)
5
checked on Jan 20, 2024
Download(s)
1
checked on Jan 20, 2024
Google ScholarTM
Check
Altmetric
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.