Multilingual Multi-class Sentiment Classification Using Convolutional Neural Networks

Multilingual Multi-class Sentiment Classification Using Convolutional Neural Networks Mohammed Attia∗, Younes Samih†, Ali Elkahky∗, Laura Kallmeyer† ∗Google Inc.; †University of D¨usseldorf New York, USA; D¨usseldorf, Germany {attia,alielkahky}@google.com {samih,kallmeyer}@phil.hhu.de Abstract This paper describes a language-independent model for multi-class sentiment analysis using a simple neural network architecture of five layers (Embedding, Conv1D, GlobalMaxPooling and two Fully-Connected). The advantage of the proposed model is that it does not rely on language-specific features such as ontologies, dictionaries, or morphological or syntactic pre-processing. Equally important, our system does not use pre-trained word2vec embeddings which can be costly to obtain and train for some languages. In this research, we also demonstrate that oversampling can be an effective approach for correcting class imbalance in the data. We evaluate our methods on three publicly available datasets for English, German and Arabic, and the results show that our system’s performance is comparable to, or even better than, the state of the art for these datasets. We make our source-code publicly available. Keywords: sentiment analysis, sentiment detection, multi-class 1. Introduction The social media has revolutionized the web by transform- ing users from being passive recipients of information into contributers and influencers. This has a direct impact on businesses, products and governance. Many of the users’ posts are opinions about products and brands that impact other consumers’ buying decisions and affect brand trust- worthiness. Negative reviews circulated online may cause critical problems for the reputation, competitive power, and survival chances of any business. This in turn has led to some fundamental changes in how businesses approach their customers, gauge satisfaction, provide support and manage risks. Sentiment analysis is formally defined as the task to iden- tify and analyze subjective information of people’s opin- ions in social media sources (Pham and Le, 2016; Yang et al., 2017). This field of study has recently attracted a lot of attention due to its implications for businesses and govern- ments. The challenges of this task can be summarized in the following points: • There is a large variety of expressions to denote a range of sentiments. Therefore a sentiment dictionary, no matter how large it could be, cannot list all the pos- sible ways people can express their attitudes. • Words change meaning depending on the context and the domain. For example, a “short cord” mostly in- dicates a negative opinion, while a “short boot time” signifies a positive one. • There could be long-distance dependencies between different constituents of a sentence, and without know- ing, for example, the scope of negation, the polarity of an adjective cannot be determined. • People do not always reveal their opinion in an explicit way, as they could be indirect, subtle or ironical. Users generally express a broad variety of sentiments with a wide range of degrees, but to simplify the task, sentiment analysis approaches have traditionally classified sentiments into either positive, negative or neutral. This paper describes our system for multilingual, multi- class sentiment classification using Convolutional Neu- ral Networks (CNNs). We evaluate our system on three …

Content truncated to 500 words