Sentiment classification is widely used for product reviews and in online social media such as forums, Twitter, and blogs. However, the problem of classifying the sentiment of user comments on news sites has not been addressed yet. News sites cover a wide range of domains including politics, sports, technology, and entertainment, in contrast to other online social sites such as forums and review sites, which are specific to a particular domain. A user associated with a news site is likely to post comments on diverse topics (e.g., politics, smartphones, and sports) or diverse entities (e.g., Obama, iPhone, or Google). Classifying the sentiment of users tied to various entities may help obtain a holistic view of their personality, which could be useful in applications such as online advertising, content personalization, and political campaign planning. In this paper, we formulate the problem of entity-specific sentiment classification of comments posted on news articles in Yahoo News and propose novel features that are specific to news comments. Experimental results show that our models outperform state-of-the-art baselines.
The success of deep learning often derives from well-chosen operational building blocks. In this work, we revise the temporal convolution operation in CNNs to better adapt it to text processing. Instead of concatenating word representations, we appeal to tensor algebra and use low-rank n-gram tensors to directly exploit interactions between words already at the convolution stage. Moreover, we extend the n-gram convolution to non-consecutive words to recognize patterns with intervening words. Through a combination of low-rank tensors, and pattern weighting, we can efficiently evaluate the resulting convolution operation via dynamic programming. We test the resulting architecture on standard sentiment classification and news categorization tasks. Our model achieves state-of-the-art performance both in terms of accuracy and training speed. For instance, we obtain 51.2% accuracy on the fine-grained sentiment classification task.
Recursive neural models, which use syntactic parse trees to recursively generate representations bottom-up, are a popular architecture. But there have not been rigorous evaluations showing for exactly which tasks this syntax-based method is appropriate. In this paper we benchmark {\bf recursive} neural models against sequential {\bf recurrent} neural models (simple recurrent and LSTM models), enforcing apples-to-apples comparison as much as possible. We investigate 4 tasks: (1) sentiment classification at the sentence level and phrase level; (2) matching questions to answer-phrases; (3) discourse parsing; (4) semantic relation extraction (e.g., {\em component-whole} between nouns). Our goal is to understand better when, and why, recursive models can outperform simpler models. We find that recursive models help mainly on tasks (like semantic relation extraction) that require associating headwords across a long distance, particularly on very long sequences. We then introduce a method for allowing recurrent models to achieve similar performance: breaking long sentences into clause-like units at punctuation and processing them separately before combining. Our results thus help understand the limitations of both classes of models, and suggest directions for improving recurrent models.
This paper introduces a new method to solve the cross-domain recognition problem. Different from the traditional domain adaption methods which rely on a global domain shift for all classes between source and target domain, the proposed method is more flexible to capture individual class variations across domains. By adopting a natural and widely used assumption -- "the data samples from the same class should lay on a low-dimensional subspace, even if they come from different domains", the proposed method circumvents the limitation of the global domain shift, and solves the cross-domain recognition by finding the compact joint subspaces of source and target domain. Specifically, given labeled samples in source domain, we construct subspaces for each of the classes. Then we construct subspaces in the target domain, called anchor subspaces, by collecting unlabeled samples that are close to each other and highly likely all fall into the same class. The corresponding class label is then assigned by minimizing a cost function which reflects the overlap and topological structure consistency between subspaces across source and target domains, and within anchor subspaces, respectively.We further combine the anchor subspaces to corresponding source subspaces to construct the compact joint subspaces. Subsequently, one-vs-rest SVM classifiers are trained in the compact joint subspaces and applied to unlabeled data in the target domain. We evaluate the proposed method on two widely used datasets: object recognition dataset for computer vision tasks, and sentiment classification dataset for natural language processing tasks. Comparison results demonstrate that the proposed method outperforms the comparison methods on both datasets.
This project addresses the problem of sentiment analysis in twitter; that is classifying tweets according to the sentiment expressed in them: positive, negative or neutral. Twitter is an online micro-blogging and social-networking platform which allows users to write short status updates of maximum length 140 characters. It is a rapidly expanding service with over 200 million registered users - out of which 100 million are active users and half of them log on twitter on a daily basis - generating nearly 250 million tweets per day. Due to this large amount of usage we hope to achieve a reflection of public sentiment by analysing the sentiments expressed in the tweets. Analysing the public sentiment is important for many applications such as firms trying to find out the response of their products in the market, predicting political elections and predicting socioeconomic phenomena like stock exchange. The aim of this project is to develop a functional classifier for accurate and automatic sentiment classification of an unknown tweet stream.
This paper covers the two approaches for sentiment analysis: i) lexicon based method; ii) machine learning method. We describe several techniques to implement these approaches and discuss how they can be adopted for sentiment classification of Twitter messages. We present a comparative study of different lexicon combinations and show that enhancing sentiment lexicons with emoticons, abbreviations and social-media slang expressions increases the accuracy of lexicon-based classification for Twitter. We discuss the importance of feature generation and feature selection processes for machine learning sentiment classification. To quantify the performance of the main sentiment analysis methods over Twitter we run these algorithms on a benchmark Twitter dataset from the SemEval-2013 competition, task 2-B. The results show that machine learning method based on SVM and Naive Bayes classifiers outperforms the lexicon method. We present a new ensemble method that uses a lexicon based sentiment score as input feature for the machine learning approach. The combined method proved to produce more precise classifications. We also show that employing a cost-sensitive classifier for highly unbalanced datasets yields an improvement of sentiment classification performance up to 7%.
This paper presents a new semi-supervised framework with convolutional neural networks (CNNs) for text categorization. Unlike the previous approaches that rely on word embeddings, our method learns embeddings of small text regions from unlabeled data for integration into a supervised CNN. The proposed scheme for embedding learning is based on the idea of two-view semi-supervised learning, which is intended to be useful for the task of interest even though the training is done on unlabeled data. Our models achieve better results than previous approaches on sentiment classification and topic classification tasks.
Emoticons (e.g., :) and :( ) have been widely used in sentiment analysis and other NLP tasks as features to ma- chine learning algorithms or as entries of sentiment lexicons. In this paper, we argue that while emoticons are strong and common signals of sentiment expression on social media the relationship between emoticons and sentiment polarity are not always clear. Thus, any algorithm that deals with sentiment polarity should take emoticons into account but extreme cau- tion should be exercised in which emoticons to depend on. First, to demonstrate the prevalence of emoticons on social media, we analyzed the frequency of emoticons in a large re- cent Twitter data set. Then we carried out four analyses to examine the relationship between emoticons and sentiment polarity as well as the contexts in which emoticons are used. The first analysis surveyed a group of participants for their perceived sentiment polarity of the most frequent emoticons. The second analysis examined clustering of words and emoti- cons to better understand the meaning conveyed by the emoti- cons. The third analysis compared the sentiment polarity of microblog posts before and after emoticons were removed from the text. The last analysis tested the hypothesis that removing emoticons from text hurts sentiment classification by training two machine learning models with and without emoticons in the text respectively. The results confirms the arguments that: 1) a few emoticons are strong and reliable signals of sentiment polarity and one should take advantage of them in any senti- ment analysis; 2) a large group of the emoticons conveys com- plicated sentiment hence they should be treated with extreme caution.
We propose an effective technique to solving review-level sentiment classification problem by using sentence-level polarity correction. Our polarity correction technique takes into account the consistency of the polarities (positive and negative) of sentences within each product review before performing the actual machine learning task. While sentences with inconsistent polarities are removed, sentences with consistent polarities are used to learn state-of-the-art classifiers. The technique achieved better results on different types of products reviews and outperforms baseline models without the correction technique. Experimental results show an average of 82% F-measure on four different product review domains.
In this article, how word embeddings can be used as features in Chinese sentiment classification is presented. Firstly, a Chinese opinion corpus is built with a million comments from hotel review websites. Then the word embeddings which represent each comment are used as input in different machine learning methods for sentiment classification, including SVM, Logistic Regression, Convolutional Neural Network (CNN) and ensemble methods. These methods get better performance compared with N-gram models using Naive Bayes (NB) and Maximum Entropy (ME). Finally, a combination of machine learning methods is proposed which presents an outstanding performance in precision, recall and F1 score. After selecting the most useful methods to construct the combinational model and testing over the corpus, the final F1 score is 0.920.
Social media data in Arabic language is becoming more and more abundant. It is a consensus that valuable information lies in social media data. Mining this data and making the process easier are gaining momentum in the industries. This paper describes an enterprise system we developed for extracting sentiment from large volumes of social data in Arabic dialects. First, we give an overview of the Big Data system for information extraction from multilingual social data from a variety of sources. Then, we focus on the Arabic sentiment analysis capability that was built on top of the system including normalizing written Arabic dialects, building sentiment lexicons, sentiment classification, and performance evaluation. Lastly, we demonstrate the value of enriching sentiment results with user profiles in understanding sentiments of a specific user group.
This paper considers Aspect-based Opinion Summarization (AOS) of reviews on particular products. To enable real applications, an AOS system needs to address two core subtasks, aspect extraction and sentiment classification. Most existing approaches to aspect extraction, which use linguistic analysis or topic modeling, are general across different products but not precise enough or suitable for particular products. Instead we take a less general but more precise scheme, directly mapping each review sentence into pre-defined aspects. To tackle aspect mapping and sentiment classification, we propose two Convolutional Neural Network (CNN) based methods, cascaded CNN and multitask CNN. Cascaded CNN contains two levels of convolutional networks. Multiple CNNs at level 1 deal with aspect mapping task, and a single CNN at level 2 deals with sentiment classification. Multitask CNN also contains multiple aspect CNNs and a sentiment CNN, but different networks share the same word embeddings. Experimental results indicate that both cascaded and multitask CNNs outperform SVM-based methods by large margins. Multitask CNN generally performs better than cascaded CNN.
Neural network models have been demonstrated to be capable of achieving remarkable performance in sentence and document modeling. Convolutional neural network (CNN) and recurrent neural network (RNN) are two mainstream architectures for such modeling tasks, which adopt totally different ways of understanding natural languages. In this work, we combine the strengths of both architectures and propose a novel and unified model called C-LSTM for sentence representation and text classification. C-LSTM utilizes CNN to extract a sequence of higher-level phrase representations, and are fed into a long short-term memory recurrent neural network (LSTM) to obtain the sentence representation. C-LSTM is able to capture both local features of phrases as well as global and temporal sentence semantics. We evaluate the proposed architecture on sentiment classification and question classification tasks. The experimental results show that the C-LSTM outperforms both CNN and LSTM and can achieve excellent performance on these tasks.
Our goal is to deploy a high-accuracy system starting with zero training examples. We consider an "on-the-job" setting, where as inputs arrive, we use real-time crowdsourcing to resolve uncertainty where needed and output our prediction when confident. As the model improves over time, the reliance on crowdsourcing queries decreases. We cast our setting as a stochastic game based on Bayesian decision theory, which allows us to balance latency, cost, and accuracy objectives in a principled way. Computing the optimal policy is intractable, so we develop an approximation based on Monte Carlo Tree Search. We tested our approach on three datasets---named-entity recognition, sentiment classification, and image classification. On the NER task we obtained more than an order of magnitude reduction in cost compared to full human annotation, while boosting performance relative to the expert provided labels. We also achieve a 8% F1 improvement over having a single human label the whole set, and a 28% F1 improvement over online learning.
Despite the success of distributional semantics, composing phrases from word vectors remains an important challenge. Several methods have been tried for benchmark tasks such as sentiment classification, including word vector averaging, matrix-vector approaches based on parsing, and on-the-fly learning of paragraph vectors. Most models usually omit stop words from the composition. Instead of such an yes-no decision, we consider several graded schemes where words are weighted according to their discriminatory relevance with respect to its use in the document (e.g., idf). Some of these methods (particularly tf-idf) are seen to result in a significant improvement in performance over prior state of the art. Further, combining such approaches into an ensemble based on alternate classifiers such as the RNN model, results in an 1.6% performance improvement on the standard IMDB movie review dataset, and a 7.01% improvement on Amazon product reviews. Since these are language free models and can be obtained in an unsupervised manner, they are of interest also for under-resourced languages such as Hindi as well and many more languages. We demonstrate the language free aspects by showing a gain of 12% for two review datasets over earlier results, and also release a new larger dataset for future testing (Singh,2015).
The goal of this paper is to investigate the connection between the performance gain that can be obtained by selftraining and the similarity between the corpora used in this approach. Self-training is a semi-supervised technique designed to increase the performance of machine learning algorithms by automatically classifying instances of a task and adding these as additional training material to the same classifier. In the context of language processing tasks, this training material is mostly an (annotated) corpus. Unfortunately self-training does not always lead to a performance increase and whether it will is largely unpredictable. We show that the similarity between corpora can be used to identify those setups for which self-training can be beneficial. We consider this research as a step in the process of developing a classifier that is able to adapt itself to each new test corpus that it is presented with.
We consider the problem of learning general-purpose, paraphrastic sentence embeddings based on supervision from the Paraphrase Database (Ganitkevitch et al., 2013). We compare six compositional architectures, evaluating them on annotated textual similarity datasets drawn both from the same distribution as the training data and from a wide range of other domains. We find that the most complex architectures, such as long short-term memory (LSTM) recurrent neural networks, perform best on the in-domain data. However, in out-of-domain scenarios, simple architectures such as word averaging vastly outperform LSTMs. Our simplest averaging model is even competitive with systems tuned for the particular tasks while also being extremely efficient and easy to use. In order to better understand how these architectures compare, we conduct further experiments on three supervised NLP tasks: sentence similarity, entailment, and sentiment classification. We again find that the word averaging models perform well for sentence similarity and entailment, outperforming LSTMs. However, on sentiment classification, we find that the LSTM performs very strongly-even recording new state-of-the-art performance on the Stanford Sentiment Treebank. We then demonstrate how to combine our pretrained sentence embeddings with these supervised tasks, using them both as a prior and as a black box feature extractor. This leads to performance rivaling the state of the art on the SICK similarity and entailment tasks. We release all of our resources to the research community with the hope that they can serve as the new baseline for further work on universal sentence embeddings.
Unsupervised methods have proven effective for discriminative tasks in a single-modality scenario. In this paper, we present a multimodal framework for learning sparse representations that can capture semantic correlation between modalities. The framework can model relationships at a higher level by forcing the shared sparse representation. In particular, we propose the use of joint dictionary learning technique for sparse coding and formulate the joint representation for concision, cross-modal representations (in case of a missing modality), and union of the cross-modal representations. Given the accelerated growth of multimodal data posted on the Web such as YouTube, Wikipedia, and Twitter, learning good multimodal features is becoming increasingly important. We show that the shared representations enabled by our framework substantially improve the classification performance under both unimodal and multimodal settings. We further show how deep architectures built on the proposed framework are effective for the case of highly nonlinear correlations between modalities. The effectiveness of our approach is demonstrated experimentally in image denoising, multimedia event detection and retrieval on the TRECVID dataset (audio-video), category classification on the Wikipedia dataset (image-text), and sentiment classification on PhotoTweet (image-text).
Despite the loss of semantic information, bag-of-ngram based methods still achieve state-of-the-art results for tasks such as sentiment classification of long movie reviews. Many document embeddings methods have been proposed to capture semantics, but they still can't outperform bag-of-ngram based methods on this task. In this paper, we modify the architecture of the recently proposed Paragraph Vector, allowing it to learn document vectors by predicting not only words, but n-gram features as well. Our model is able to capture both semantics and word order in documents while keeping the expressive power of learned vectors. Experimental results on IMDB movie review dataset shows that our model outperforms previous deep learning models and bag-of-ngram based models due to the above advantages. More robust results are also obtained when our model is combined with other models. The source code of our model will be also published together with this paper.
Sentiment analysis on social media data such as tweets and weibo has become a very important and challenging task. Due to the intrinsic properties of such data, tweets are short, noisy, and of divergent topics, and sentiment classification on these data requires to modeling various contexts such as the retweet/reply history of a tweet, and the social context about authors and relationships. While few prior study has approached the issue of modeling contexts in tweet, this paper proposes to use a hierarchical LSTM to model rich contexts in tweet, particularly long-range context. Experimental results show that contexts can help us to perform sentiment classification remarkably better.