When training tweet sentiment classifiers, many data quality challenges must be addressed. One potential issue is class imbalance, where most instances belong to a single majority class. This may negatively impact classifier performance as classifiers trained on imbalanced data may favor classification of new, unseen instances as belonging to the majority class. This issue is accompanied by a second challenge, high-dimesionality, since very large numbers of text based features are used to describe tweet datasets. For datasets where both of these challenges are present, we can combine feature selection and data sampling to address both highdimensionality and class imbalance. However, three potential approaches exist for combining data sampling and feature selection and it is unclear which approach is optimal. In this paper, we seek to determine if there is a best approach for combining data sampling and feature selection. We conduct tests using random undersampling with two post-sampling class ratios (50:50 and 35:65) combined with three feature rankers. Classifiers are trained with each potential combination approach using seven different learners on two datasets. We found that, overall, classifiers trained by performing feature selection followed by data sampling performed better than the other two approaches; however, the differences were only significant for the more imbalanced dataset.

Sentiment analysis became a hot topic, specially with the amount of opinions available in social media data. With the increasing interest in this theme, several methods have been proposed in the literature. Recent efforts have showed that there is no single method that always achieves the best prediction performance for different datasets. Additionally, novel methods have not being extensively compared with other methods and across different datasets, specially methods that are not designed to the English language. Consequently, researchers tend to accept any popular method as a valid methodology to measure sentiments, a practice that is usual in science. In this context, we propose iFeel 2.0, an online web system that implements 19 sentence-level sentiment analysis methods and allows users to easily label a dataset with all of them. iFeel aims at easing the comparison of new methods with baseline approaches and can also be helpful for those interested in using sentiment analysis, allowing them to choose an appropriate sentiment analysis method that works fine for a new dataset. We also incorporate a multiple language feature to allow methods designed for specific languages to be easily compared with a baseline approach that simply translates the input data to English and run these 19 methods. We hope this system can represent an important contribution to this field. Sentiment analysis became a hot topic, specially with the amount of opinions available in social media data.With the increasing interest in this theme, several methods have been proposed in the literature. Recent effortshave showed that there is no single method that always achieves the best prediction performance for different datasets. Additionally, novel methods have not being extensively compared with other methods and across different datasets, specially methods that are not designed to the English language.Consequently, researchers tend to accept any popular method as a valid methodology to measure sentiments, a practice that is usual in science.In this context, we propose iFeel 2.0, an online web system that implements 19 sentence-level sentiment analysis methods and allows users to easily label a dataset with all of them. iFeel aims at easing the comparison of new methods with baseline approaches and can also be helpful for those interested in using sentiment analysis, allowing them to choose an appropriate sentiment analysis method that works fine for a new dataset.We also incorporate a multiple language feature to allow methods designed for specific languages to be easily compared with a baseline approach that simply translates the input data to English and run these 19 methods. We hope this system can represent an important contribution to this field.

The tracking of citizens' reactions in social media during crises has attracted an increasing level of interest in the research community. In particular, sentiment analysis over social media posts can be regarded as a particularly useful tool, enabling civil protection and law enforcement agencies to more effectively respond during this type of situation. Prior work on sentiment analysis in social media during crises has applied well-known techniques for overall sentiment detection in posts. However, we argue that sentiment analysis of the overall post might not always be suitable, as it may miss the presence of more targeted sentiments, e.g. about the people and organizations involved (which we refer to as sentiment targets). Through a crowdsourcing study, we show that there are marked differences between the overall tweet sentiment and the sentiment expressed towards the subjects mentioned in tweets related to three crises events.

Social constraints have replaced the natural cycle of light and darkness as the main determinant of wake-up and activity times for many people. In this paper we show how Twitter activity can be used as a source of large-scale, naturally occurring data for the study of circadian rhythm in humans. Our year-long initial study is based on almost 1.5 million observations by over 200,000 users. The progression of the onset of Twitter activity times on free days in the course of the year is consistent with previous survey-based research on wake times. We show that the difference in wake-up time (implicating lack of sleep) on weekdays compared to Sundays is between 1 hour and over 2 hours depending on the time of year. The data also supports the assertion that Daylight Saving Time greatly disrupts the easing of social jetlag in the Spring transition.

This paper presents a novel approach to perform sentiment analysis of news videos, based on the fusion of audio, textual and visual clues extracted from their contents. The proposed approach aims at contributing to the semiodiscoursive study regarding the construction of the ethos (identity) of this media universe, which has become a central part of the modern-day lives of millions of people. To achieve this goal, we apply state-of-the-art computational methods for (1) automatic emotion recognition from facial expressions, (2) extraction of modulations in the participants' speeches and (3) sentiment analysis from the closed caption associated to the videos of interest. More specifically, we compute features, such as, visual intensities of recognized emotions, field sizes of participants, voicing probability, sound loudness, speech fundamental frequencies and the sentiment scores (polarities) from text sentences in the closed caption. Experimental results with a dataset containing 520 annotated news videos from three Brazilian and one American popular TV newscasts show that our approach achieves an accuracy of up to 84% in the sentiments (tension levels) classification task, thus demonstrating its high potential to be used by media analysts in several applications, especially, in the journalistic domain.

This study analyzes different methodological approaches followed in social media literature and their accuracy in predicting the general elections of four countries. Volumetric and unsupervised and supervised sentiment approaches are adopted for generating 12 metrics to compute predicted voteshares. The findings suggest that Twitter-based predictions can produce accurate results for elections, given the digital environment of a country. A cross-country analyses helps to evaluate the quality of predictions and the influence of different contexts, such as technological development and democratic setups. We recommend future scholars to combine volume, sentiment and network aspects of social media to model voting intentions in developing societies.

An overwhelming amount of data is generated everyday onsocial media, encompassing a wide spectrum of topics. With almost every business decision depending on customer opinion, mining of social media data needs to be quick and easy.For a data analyst to keep up with the agility and the scale of the data, it is impossible to bank on fully supervised techniques to mine topics and their associated sentiments from social media. Motivated by this, we propose a weakly supervised approach (named, TweetGrep) that lets the data analyst easily define a topic by few keywords and adapt a generic sentiment classifier to the topic – by jointly modeling topics and sentiments using label regularization. Experiments with diverse datasets show that TweetGrep beats the state-of-the-art models for both the tasks of retrieving topical tweet sand analyzing the sentiment of the tweets (average improvement of 4.97% and 6.91% respectively in terms of area under the curve). Further, we show that TweetGrep can also be adopted in a novel task of hashtag disambiguation, which significantly outperforms the baseline methods.

Is it time to bring text analytics in-house? A director of analytics at Visa, who is slated to speak at the 13th Annual Text Analytics Summit West in San Francisco next week, advises CIOs to consider these three questions before making a move. Text analytics and the CIO don't ordinarily go hand-in-hand. Instead, the emerging field of text analytics, or the mining of text to derive business insights, is typically farmed out to experts. But if the business is serious about investing in tools and bringing the technology in-house, CIOs should be involved, according to Ramkumar Ravichandran, a director of analytics at Visa Inc.

Earlier this year I wrote a post that showed how to perform sentiment analysis in Dynamics CRM using Microsoft Azure Text Analytics. Azure Text Analytics makes it incredibly easy to use sentiment analysis (with English text only), but the full Azure Machine Learning offering is much more powerful. In today's post I will show how to create a custom predictive web service in Azure ML and make predictions with it in Dynamics CRM. One of the exciting announcements about Dynamics CRM 2016 is that it includes some sort of integration with Azure ML, so what's the point of this blog post? For this demonstration I am using data from the AdventureWorks data warehouse sample database to build a model to predict whether a contact in CRM is likely to be a bicycle buyer.

A lot of discussion around Matt Jockers' Syuzhet package (involving Annie Swafford, Ted Underwood, Andrew Piper, Scott Weingart and many others) has focused on issues of validity -- whether sentiment analysis is accurate enough for the task, whether the Fourier transform is an appropriate method for dimensionality reduction, whether the emotional trajectories themselves are valid measurements of anything at all (Scott has a good enumeration of the various issues here.) Andrew's discussion of the validity of inherently subjective measurements inspired me to solicit at least one data point from readers that we can use for one question under discussion with Syuzhet: what does a human judgment of the "emotional trajectory" of a work look like, and how often do readers agree with each other on this task? This method of soliciting human judgments for inherently subjective tasks is at the core of NLP and a lot of machine learning -- syntactic parsing, part of speech tagging, named entity recognition, topic classification, sentiment analysis, and lots of other tasks all rely on humans making judgments that are often surprisingly difficult in practice; learning algorithms in these cases are not so much learning any notion of "truth" but simply to reproduce the human judgments they're given. Agreement rates between humans is often seen as a proxy for the complexity of the task; if humans can't agree, it can be a sign that the task is ill-defined or underspecified. Word sense disambiguation is one good example of this, with low inter-annotator agreement rates [Snyder and Palmer 2004]; while sentiment analysis was originally designed with product/movie reviews in mind (does person X like product Y?) -- i.e., attitude with respect to a particular target -- I think the more general sentiment-as-tone problem (is this tweet happy or sad?) is much less well specified as a problem with an answer that can be judged by anyone but the original author. One aspect of those kind of annotations that I think is much less explored (which Piper points to and I think would be an extremely interesting area to work on) is the case where multiple judgments are simultaneously valid -- different interpretations of the same phenomenon, each backed by their own argument.