THU_NGN at SemEval-2018 Task 10: Capturing Discriminative Attributes with MLP-CNN model

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 958–962 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics THU NGN at SemEval-2018 Task 10: Capturing Discriminative Attributes with MLP-CNN model Chuhan Wu1, Fangzhao Wu2, Zhigang Yuan1, Sixing Wu1 and Yongfeng Huang1 1Tsinghua National Laboratory for Information Science and Technology, Department of Electronic Engineering, Tsinghua University Beijing 100084, China 2Microsoft Research Asia {wuch15,yuanzg14,wu-sx15,yfhuang}@mails.tsinghua.edu.cn wufangzhao@gmail.com Abstract Existing semantic models are capable of iden- tifying the semantic similarity of words. How- ever, it’s hard for these models to discrimi- nate between a word and another similar word. Thus, the aim of SemEval-2018 Task 10 is to predict whether a word is a discrimina- tive attribute between two concepts. In this task, we apply a multilayer perceptron (MLP)- convolutional neural network (CNN) model to identify whether an attribute is discriminative. The CNNs are used to extract low-level fea- tures from the inputs. The MLP takes both the flatten CNN maps and inputs to predict the la- bels. The evaluation F-score of our system on the test set is 0.629 (ranked 15th), which in- dicates that our system still needs to be im- proved. However, the behaviours of our sys- tem in our experiments provide useful infor- mation, which can help to improve the collec- tive understanding of this novel task. 1 Introduction Evaluating the similarity of words is an important task in semantic modeling. There have been dif- ferent approaches based on corpus statistics (Jiang and Conrath, 1997; Mihalcea et al., 2006) and on- tology (Seco et al., 2004; S´anchez et al., 2012). After an effective word representation proposed by mikolov et al (2013), word similarity can be evalu- ated based on word embedding weights (Levy and Goldberg, 2014). Usually higher cosine similar- ity of word embedding vectors indicates higher se- mantic similarity. However, existing semantic methods are not capable of discriminating similar words between each other without additional information. For ex- ample, it is easy for these models to tell “dog” and “puppy” is similar, but they can’t tell the differ- ences between each other. It limits the use of these models to mine such fine-grained semantic infor- mation from texts. Thus, the SemEval-2018 Task 10 is proposed to determine whether an attribute can help to discriminate between two words(Krebs et al., 2018). One can express semantic differences between concepts by referring to attributes associ- ated with those concepts. The differences between concepts can usually be identified by the presence or absence of specific attributes. For example, the attributes “red” and “yellow” are discriminative for concepts “apple” and “banana”, while “sweet” or “fruit” are not discriminative. Capturing such discriminative attributes can be regarded as a binary classification task: given two words and an attribute, predict whether the at- tribute is a difference between the two words. Ex- isting methods to capture discriminative attributes are mainly based on dictionary (Parikh and Grau- man, 2011). In recent years, CNN have been suc- cessfully applied to text classification task (Kim, 2014). …

Content truncated to 500 words