funSentiment at SemEval-2017 Task 4: Topic-Based Message Sentiment Classification by Exploiting Word Embeddings, Text Features and Target Contexts

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 741–746, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics funSentiment at SemEval-2017 Task 4: Topic-Based Message Senti- ment Classification by Exploiting Word Embeddings, Text Features and Target Contexts Quanzhi Li, Armineh Nourbakhsh, Xiaomo Liu, Rui Fang, Sameena Shah Research and Development Thomson Reuters 3 Times Square, NYC, NY 10036 {quanzhi.li, armineh.nourbakhsh, xiaomo.liu, rui.fang, sameena.shah}@thomsonreuters.com Abstract This paper describes the approach we used for SemEval-2017 Task 4: Sentiment Analysis in Twitter. Topic-based (target-dependent) sentiment analysis has become attractive and been used in some applications recently, but it is still a challenging research task. In our approach, we take the left and right context of a target into consideration when generating polarity classification features. We use two types of word embeddings in our classifiers: the general word embeddings learned from 200 million tweets, and sen- timent-specific word embeddings learned from 10 million tweets using distance supervision. We also incorporate a text feature model in our algorithm. This model produces features based on text negation, tf.idf weighting scheme, and a Rocchio text classification method. We participated in four subtasks (B, C, D & E for English), all of which are about topic-based message polarity classification. Our team is ranked #6 in subtask B, #3 by MAEu and #9 by MAEm in sub- task C, #3 using RAE and #6 using KLD in subtask D, and #3 in subtask E. 1 Introduction There have been many studies on message or sentence level sentiment classification (Go et al., 2009; Mohammand et al., 2013; Pang et al., 2002; Liu, 2012; Tang et al., 2014), but there are few studies on target-dependent, or topic-based, sentiment prediction (Jiang et al., 2011; Dong et al., 2014; Vo and Zhang, 2015). A target entity in a message does not necessarily have the same polarity type as the message, and different enti- ties in the same message may have different po- larities. For example, in the tweet “Linux is bet- ter than Windows”, the two named entities, Linux and Windows, will have different senti- ment polarities. In this paper, we describe our approach for the subtask B, C, D & E of SemEval-17 Task 4: Sentiment Analysis in Twit- ter (Sara Rosenthal and Noura Farra and Preslav Nakov, 2017). All the four subtasks are on top- ic-based message sentiment classification. Task B and C are about topic-based message polarity classification. Given a message and a topic, in task B, we classify the message on a two-point scale: positive or negative sentiment towards the topic. And in task C, we classify the message on a five-point scale: sentiment conveyed by the tweet towards the topic on a five-point scale. Task D and E are about Tweet quantification. Given a set of tweets about a given topic, in task D, we want to estimate the distribution of the tweets across two-point scale - the positive and negative classes, and in task E, we estimate that on a five-point …

Content truncated to 500 words