|
|
 |
| |
|
|
|
A Multi-Perspective Knowledge Discovery Approach for Word Sense Disambiguation |
|
|
|
PP: 61-71 |
|
|
|
doi:10.18576/amis/13S107
|
|
|
|
Author(s) |
|
|
|
Rajini. S,
Vasuki. A,
|
|
|
|
Abstract |
|
|
| In this paper, a multi-perspective knowledge discovery approach for word sense disambiguation is proposed. Initially a two-step pre-processing is carried out which includes stop word removal and stemming. From the context of an ambiguous word, the features are extracted such as word embedding, continuous bag-of-words and skip gram models. The unigrams and bigrams are extracted from the text and the bigrams are integrated with the unigrams. Then distributional similarity and semantic similarity scores are evaluated based on the local mutual information, point-wise local mutual information and the feature values. For the context classification, convolutional neural network model is utilized. In order to get strong baseline result, the distributional similarity and the semantic similarity matching algorithm is applied for the text features, particularly the unigram representation process. SemEval-2010 Word Sense Induction and Disambiguation dataset is used in this work. The experimental analysis is carried out by implementing various classifiers such as KNN, Na ̈ıve Bayes and Random Forest methods. The proposed approach provides good outcomes in terms of accuracy, F-measure, precision and recall.
|
|
|
|
|
|
|
|
