Topic-Based Document-Level Sentiment Analysis Using Contextual Cues

Abstract

Document-level Sentiment Analysis is a complex task that implies the analysis of large textual content that can incorporate multiple contradictory polarities at the phrase and word levels. Most of the current approaches either represent textual data using pre-trained word embeddings without considering the local context that can be extracted from the dataset, or they detect the overall topic polarity without considering both the local and global context. In this paper, we propose a novel document-topic embedding model, DocTopic2Vec, for document-level polarity detection in large texts by employing general and specific contextual cues obtained through the use of document embeddings (Doc2Vec) and Topic Modeling. In our approach, (1) we use a large dataset with game reviews to create different word embeddings by applying Word2Vec, FastText, and GloVe, (2) we create Doc2Vecs enriched with the local context given by the word embeddings for each review, (3) we construct topic embeddings Topic2Vec using three Topic Modeling algorithms, i.e., LDA, NMF, and LSI, to enhance the global context of the Sentiment Analysis task, (4) for each document and its dominant topic, we build the new DocTopic2Vec by concatenating the Doc2Vec with the Topic2Vec created with the same word embedding. We also design six new Convolutional-based (Bidirectional) Recurrent Deep Neural Network Architectures that show promising results for this task. The proposed DocTopic2Vecs are used to benchmark multiple Machine and Deep Learning models, i.e., a Logistic Regression model, used as a baseline, and 18 Deep Neural Networks Architectures. The experimental results show that the new embedding and the new Deep Neural Network Architectures achieve better results than the baseline, i.e., Logistic Regression and Doc2Vec.

Mining Public Opinions on COVID-19 Vaccination: A Temporal Analysis to Support Combating Misinformation

“…The logistic regression (LogReg) classifier is a linear regression algorithm to make predictions when the dependent variable is binary [ 35 ]. The algorithm uses a function to minimize the estimators’ errors using the log-likelihood and applies gradient descent to determine the parameters that produce the best estimators [ 36 ]. For the classification task, according to Mitroi et al [ 37 ], given a set of documents d i , where y i is the class of a document, de logistic regression maps the documents to classes using a sigmoid function h θ ( d i ) to determine the parameters of the vector fitting the regression line describer by Equation (2): …”

Public Attention and Sentiment toward Intimate Partner Violence Based on Weibo in China: A Text Mining Approach

“…When using the method of sentiment analysis, many researchers used different analysis methods. Zhang et al [ 56 ], Shakil et al [ 57 ] and Phu et al [ 58 ] used dictionary-based methods; Truică et al [ 59 ], Messina et al [ 60 ], Wang et al [ 61 ] and other studies used machine learning, whereas Szabóová et al [ 62 ], Ahmed et al [ 63 ] combine the dictionary-based methods and machine learning. The sentiment dictionary-based method is good at processing fine-grained text sentiment analysis, which is conducive to analyzing sentiment characteristics in specific fields [ 64 ].…”

Navigating the Digital Neurolandscape: Analyzing the Social Perception of and Sentiments Regarding Neurological Disorders through Topic Modeling and Unsupervised Research Using Twitter

“…In contrast, leveraging LDA necessitates only two steps: preprocessing and applying the LDA model. Also, despite our use of the distilROBERTa database, topic modeling techniques like proper LDA could be used to deepen the sentiment analysis, as has been performed in other works [60]. The use of information diffusion models, such as the MABED (Modeling and Analysis of Burst Event Dynamics) or Peaky Topics algorithm, to study the temporal evolution of topics has also been considered in other works [61].…”