EoANN: Lexical Semantic Relation Classification Using an Ensemble of Artificial Neural Networks

Proceedings of Recent Advances in Natural Language Processing, pages 481–486, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_057 481 1 Abstract Researchers use wordnets as a knowledge base in many natural language processing tasks and applications, such as question answering, textual entailment, discourse classification, and so forth. Lexical semantic relations among words or concepts are important parts of knowledge encoded in wordnets. As the use of wordnets becomes extensively widespread, extending the existing ones gets more attention. Manual construction and extension of lexical semantic relations for WordNets or knowledge graphs are very time consuming. Using automatic relation extraction methods can speedup this process. In this study, we exploit an ensemble of LSTM and convolutional neural networks in a supervised manner to capture lexical semantic relations which can either be used directly in NLP applications or compose the edges of wordnets. The whole procedure of learning vector space representation of relations is language independent. We used Princeton WordNet 3.1, and FarsNet 3.0 (the Persian wordnet), as gold standards to evaluate the predictive performance of our model and the results are comparable on the two languages. Empirical results demonstrate that our model outperforms the state-of- the-art models. 1 Introduction Lexical semantic relation classification is the task of identifying s semantic relation(s) which holds between word pairs among a set of predefined relation types. Relation classification can be done in a supervised manner, using a dataset, labeled with a certain number of relation classes. In addition to classification with known relations, there are some methods which go even further and learn new semantic relations and suggest new relation categories (Shamsfard and Barforoosh, 2003). Relation identification plays an essential role in many natural language processing application such as question answering, recognizing textual entailment and discourse understanding. There are two main approaches for classification of lexical semantic relations; distributional and path-based (Wang et al., 2017). Path-based approaches try to recognize the type of semantic relation between word pairs according to their co-occurrence information in the corpus. These methods mainly use the dependency path between word pairs as their input feature (Snow et al., 2004; Riedel et al., 2013). As Ziph's law states that most of the words in vocabulary rarely occur in the corpus (Powers, 1998) these methods have some limitation for word pairs who do not co-occur in a context. On the other hand according to the distributional hypothesis which states "words that occur in similar contexts tend to have similar meanings" (Harris, 1954), distributional approaches try to recognize the relation between words based on their separate occurrence in the corpus which can EoANN: Lexical Semantic Relation Classification Using an Ensemble of Artificial Neural Networks Rayehe Hosseinipour, Mehrnoosh Shamsfard Computer Engineering Department, Shahid Beheshti university, Tehran, Iran Computer Engineering Department, Shahid Beheshti university, Tehran, Iran r.hosseinipour@mail.sbu.ac.ir m-shams@sbu.ac.ir 482 2 be represented for example by their word embedding vectors (Mikolov et al.,2013) and these methods have shown great performance (Baroni et al., 2012; Turney and Pantel, 2010; Roller et al., 2014). In the last decades, several researches …

Content truncated to 500 words