Sentiment analysis on social media relies on comprehending the natural language and using a robust machine learning technique that learns multiple layers of representations or features of the data and produces state-of-the-art prediction results. The cultural miscellanies, geographically limited trending topic hash-tags, access to aboriginal language keyboards, and conversational comfort in native language compound the linguistic challenges of sentiment analysis. This research evaluates the performance of cross-lingual contextual word embeddings and zero-shot transfer learning in projecting predictions from resource-rich English to resource-poor Hindi language. The cross-lingual XLM-RoBERTa classification model is trained and fine-tuned using the English language Benchmark SemEval 2017 dataset Task 4 A and subsequently zero-shot transfer learning is used to evaluate the classification model on two Hindi sentence-level sentiment analysis datasets, namely, IITP-Movie and IITP-Product review datasets. The proposed model compares favorably to state-of-the-art approaches and gives an effective solution to sentence-level (tweet-level) analysis of sentiments in a resource-poor scenario. The proposed model compares favorably to state-of-the-art approaches and achieves an average performance accuracy of 60.93 on both the Hindi datasets.

Sentiment Analysis Using XLM-R Transformer and Zero-shot Transfer Learning on Resource-poor Indian Language

Text classification is a basic task in natural language processing. When the amount of data is insufficient, the classification accuracy will be greatly affected. We propose to use the back-translation method to expand three Chinese data sets used for text classification, and then train and predict the data sets through deep learning classification model. The results prove that using back-translation to expand the data is particularly helpful on a smaller dataset, it also can reduce the unbalanced distribution of samples and improve the classification performance.

https://iopscience.iop.org/article/10.1088/1742-6596/1651/1/012039/pdf

Data Augmentation For Chinese Text Classification Using Back-Translation

In this paper, we discuss an approach to sentiment analysis and emotion identification for user comments. The solution is threefold: 1) topic detection, 2) sentiment evaluation, 3) toxicity detection and toxic spans localization. The lack of significantly large training data for the Russian language is handled by utilizing multilingual word embeddings, the adversarial domain adaptation model, and data augmentation. We present an overview of various preprocessing pipelines for topic modeling and highlight the LDA- Mallet model which demonstrates the best performance. For sentiment analysis and toxicity detection, we examine the efficacy of a support vector machine and a deep neural network with a multilingual language model and adversarial domain adaptation that allows us to train algorithms with datasets in the English language. All methods are tested with a dataset of user comments to various online-courses and adjusted to provide support for the development of a virtual dialogue assistant for conducting virtual exams.

Multilingual Sentiment Analysis and Toxicity Detection for Text Messages in Russian

The information age poses new challenges to the automatic and rapid classification of public security data. To get rid of the low-efficiency work mode of manual identification, perform rapid automatic classification of police situation, and improve the performance and accuracy of classifiers, many models and algorithms have been proposed. The traditional police situation classification methods almost always use traditional machine learning algorithms. The shortcomings of complex working mode, heavy workload, and poor learning effect have led to the rise of deep learning classification methods. This paper proposes the ATextCNN model, which introduces an attention mechanism in the Input Layer based on the TextCNN model, and uses a word filtering algorithm to calculate the contribution of words to filter nonsense words. The ATextCNN model develops a multi-classification model based on deep learning. We conduct experiments on two real data sets and divided data into nine categories. Experiments show that the ATextCNN model is better than the traditional classification methods in terms of accuracy and efficiency when performing multi-classification tasks, and is better than the state of the art method in practical application. Therefore, our method is more conducive to police situation identification and management.

ATextCNN Model: A New Multi-classification Method for Police Situation.

One of the purposes of detecting the crisis related tweets is the ability to single out the tweets that provide information about the helps needed and offered. Classification of such tweets is difficult because of the unavailability of sufficient annotated tweets in those categories. To facilitate such classifications, a domain and event adaptive augmentation approach is proposed. The main objective of the research is to enhance the classification of crisis related tweets that have less training samples. The proposed algorithms are designed to integrate the innate domain and event information during the selection of words for augmentation. Components such as CrisisLex lexicon, Word2Vec embeddings and WordNet are utilized for the proposed augmentation. Experimentation is carried out to substantiate the benefits of augmentation. Results indicate increased performance of the classifier when provided with the expanded dataset including the augmented and original tweets. To combat the problem of overfitting and class imbalance arising due to the lesser training samples, a novel tweets augmentation algorithm can be utilized. The advantage in the proposed algorithms is the ability to retain the structure and inherent nature of the tweets during the augmentation.

A novel domain and event adaptive tweet augmentation approach for enhancing the classification of crisis related tweets

Introducing negative items into sentences can shift the polarity of emotional words and leads to misclassification. Therefore, dealing with the negative item is indispensable to the analysis of the polarity of tweets. This paper first uses the combination of Conjunction Analysis (CA) technology and Punctuation Mark Identification (PMI) technology to detect negation cue and its scope. Besides, we propose the OL-DAWE model, which uses Data Augmentation(DA) technology to generate opposed tweets according to the original tweet. The model extends the training data set, and test data set and learns the original and opposed sides of the tweet in the training module. When predicting the polarity of tweets, the OL-DAWE model considers the positive degree (negative degree) of the original tweet and the negative degree (positive degree) of its opposed tweet. We conduct experiments on two real-world data sets. We prove the effectiveness of our combined technology in negation processing and show that the OL-DAWE model in the polarity sentiment analysis of tweets is better than the baseline for its simplicity and high efficiency.

/pdf/the-ol-dawe-model-tweet-polarity-sentiment-analysis-with-4imn1f89nj.pdf

The OL-DAWE Model: Tweet Polarity Sentiment Analysis With Data Augmentation

Twitter produces various colloquial tweets as an open communication platform. Previous research shows the frequency of negative sentences in spoken sentences is twice that of written texts. Negative items in negative sentences can shift the polarity of words with feelings, and leads to wrong classification. Therefore, negation processing is essential for the sentiment classification of tweets generated by Twitter. On the basis of considering the importance of negation which is often ignored in previous work, this paper firstly combines the technique of Conjunction Analysis (CA) with the technique of Punctuation Mark Identification (PMI) to detect the negation clue and its scope more accurately. In addition, We propose the OL-DAWE model. The model uses Data Augmentation (DA) approach to generate the opposed tweet according to the original tweet. The model extends learnable data and learns its polarity from both of positive and negative aspects of a tweet. In predicting the polarity of a tweet, the OL-DAWE model takes the positive (negative) degree of the original tweet into account, and also considers the negative (positive) degree of the opposed tweet. We conduct two experiments on two real-world data sets and analyze the experimental results from the perspectives of accuracy and robustness. We prove the effectiveness of our combined technology in negation processing and show that our OL-DAWE model in the polarity sentiment analysis of tweets is better than the baseline for its simplicity and high efficiency.

Ding Feng

Papers

The OL-DAWE Model: Tweet Polarity Sentiment Analysis With Data Augmentation

ATextCNN Model: A New Multi-classification Method for Police Situation.

Data Augmentation for Polarity Sentiment Analysis