Using Random Undersampling to Alleviate Class Imbalance on Tweet Sentiment Data

Prusa, Joseph D.; Khoshgoftaar, Taghi M.; Dittman, David J.; Napolitano, Amri

doi:10.1109/iri.2015.39

Cited by 117 publications

(56 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Interestingly, they found random under-sampling did make significant improvement compared with not using any technique at all, which is quite different from the conclusion drawn in [49].…”

Section: Single Classifier Based Approachescontrasting

confidence: 65%

“…From our observation of the data we acquire via Twitter API, the number of tweets with positive emojis is always greater than the amount with negative Emojis (from the raw data we draw, the size of positive ones could double the negative ones). From paper [49], [91], imbalanced data could introduce some potential problem on the classification result.…”

Section: Data Resamplingmentioning

confidence: 99%

See 1 more Smart Citation

A comprehensive topic-model based hybrid sentiment analysis system

Yang¹

View full text Add to dashboard Cite

Nowadays, Twitter sentiment analysis is drawing a lot of attention due to its potential to drive decision making in a variety of domains. However, the trend that publicly available training datasets are becoming less available, the difficulty in determining topic numbers for topic model-based approach, and a lack of data level discussion about how to utilize the proposed models day-today to drive applications are still the remained concerns. To solve these problems, we firstly offer a new method to collect and build Twitter training dataset based on noisy labels; In addition, we proposed a topic-model based hybrid sentiment classification model by using our self-collected tweets, which utilizes three different topic models and coherence score to choose the best topic model in an automated way; Last but not least, a use case is illustrated to show how to apply our pipeline in a daily basis to solve real business problems.

show abstract

“…Interestingly, they found random under-sampling did make significant improvement compared with not using any technique at all, which is quite different from the conclusion drawn in [49].…”

Section: Single Classifier Based Approachescontrasting

confidence: 65%

Section: Data Resamplingmentioning

confidence: 99%

A comprehensive topic-model based hybrid sentiment analysis system

Yang¹

View full text Add to dashboard Cite

show abstract

“…Random undersampling is a method of data sampling which randomly selects most of the class instances and removes them until the desired class distribution is attained [14].…”

Section: Samplingmentioning

confidence: 99%

Effectiveness Undersampling Method and Feature Reduction in Credit Card Fraud Detection

Trisanto¹,

Rismawati²,

Mulya³

et al. 2020

IJIES

View full text Add to dashboard Cite

Credit card fraud is an issue that has affected Indonesia payment system over a decade. Sometimes, the result of the fraud used for terrorism and other crimes. Financial loss is not the only problem that is affected caused by credit card fraud but also Indonesia images in international trade, e-commerce, and the merchant. Currently, a trusted and secured banking payment system is crucial for both customers and banks. The problem from credit card fraud dataset is the data have many features and imbalanced class, this problem leads the paper to propose undersampling technique and feature reduction methods. In this paper we proposed two stage-feature reduction technique because a stage feature reduction could not find the optimal features. On the other hands, we are also applied Instance Hardness Threshold sampling and Random undersampling to deal with imbalance data. The two-stage feature reduction is chosen to eliminate the ineffective feature that cannot eliminate using only one feature reduction. The model from the implemented machine learning methods is evaluated using accuracy, specificity, recall, and Matthews Correlation Coefficient. We implemented our proposed approaches in the ULB credit card fraud detection dataset. According to the result, the undersampling gives a boost in performances which improve the recall and MCC score, the IHT undersampling provide goods results, and in some cases, the result can predict all the test set correctly. However, the two-stages feature reduction fails to improve the accuracy, precision, recall, and MCC score. In one case, the method reduced the accuracy score to 0.302.

show abstract

“…The minority class is not modified during this procedure. The random under-sampling technique has been shown to improve model classification performance, as compared to training on the entire dataset without RUS [47,49].…”

Section: Random Under-sampling (Rus)mentioning

confidence: 99%

A parallel and distributed stochastic gradient descent implementation using commodity clusters

Kennedy

Khoshgoftaar

Villanustre³

et al. 2019

J Big Data

Self Cite

View full text Add to dashboard Cite

IntroductionTraining neural networks effectively and efficiently is an important component of Deep Learning. Large neural networks can consist of dozens, hundreds or even thousands of layers each with thousands of artificial neurons. Depending on the network's architecture, each of these neurons is connected to a large number of other neurons, where each connection has a trainable weight parameter that determines how the network responds to input signals. In the context of this paper, the effective training of these large complex networks is accomplished through the use of the computationally expensive process of backpropagation. Additionally, neural networks benefit from training on Big Data, as typically more data produces more performant models [1]. For example, the ImageNet database AlexNet was trained on roughly 1.2 million images, and at the time achieved state of the art results [2]. Problems of this magnitude are common and thus researching parallel network optimization on distributed and parallel systems is highly important. AbstractDeep Learning is an increasingly important subdomain of artificial intelligence, which benefits from training on Big Data. The size and complexity of the model combined with the size of the training dataset makes the training process very computationally and temporally expensive. Accelerating the training process of Deep Learning using cluster computers faces many challenges ranging from distributed optimizers to the large communication overhead specific to systems with off the shelf networking components. In this paper, we present a novel distributed and parallel implementation of stochastic gradient descent (SGD) on a distributed cluster of commodity computers. We use high-performance computing cluster (HPCC) systems as the underlying cluster environment for the implementation. We overview how the HPCC systems platform provides the environment for distributed and parallel Deep Learning, how it provides a facility to work with third party open source libraries such as TensorFlow, and detail our use of third-party libraries and HPCC functionality for implementation. We provide experimental results that validate our work and show that our implementation can scale with respect to both dataset size and the number of compute nodes in the cluster.

show abstract

Using Random Undersampling to Alleviate Class Imbalance on Tweet Sentiment Data

Cited by 117 publications

References 9 publications

A comprehensive topic-model based hybrid sentiment analysis system

A comprehensive topic-model based hybrid sentiment analysis system

Effectiveness Undersampling Method and Feature Reduction in Credit Card Fraud Detection

A parallel and distributed stochastic gradient descent implementation using commodity clusters

Contact Info

Product

Resources

About