Sentiment Representation Research Articles

Modeling group-level public sentiment in social networks through topic and role enhancement

Public sentiment within social networks exerts a profound influence on societal dynamics, underscoring the increasing demand for accurate public opinion prediction. Most existing methods predominantly measure sentiment by quantifying user sentiments individually, overlooking group-level factors that crucially contribute to public sentiment. Thus, based on our finding that public sentiment is primarily shaped by user-group interactions and their interplay with evolving topics, we innovatively model the forming process of public sentiment at the group level. In this paper, we propose the Topic and Role Enhanced Group-level Public Sentiment Prediction model (TRESP), capturing the intricate interplay among sentiment, topic, and role. Specifically, an LSTM neural network is firstly leveraged to trace the temporal correlations between topics and sentiment shifts, yielding a topic-informed content sentiment representation. Subsequently, a specially crafted hierarchical attention network captures social and role attributes, representing the overarching social group environment. Finally, we predict future public sentiment by merging the derived group sentiment representation with the group social representation, demonstrating a holistic insight into the sentiment trajectory. Extensive experiments were conducted on two real-world datasets of over 30,000 tweets collected from more than 14,000 users to validate our model. Notably, our model significantly outperforms the state-of-the-art approaches in public sentiment prediction, indicating the importance and effectiveness of encapsulating interactions both within and among user subgroups.

EMFSA: Emoji-based multifeature fusion sentiment analysis

Short texts on social platforms often suffer from insufficient emotional semantic expressions, sparse features, and polysemy. To enhance the accuracy achieved by sentiment analysis for short texts, this paper proposes an emoji-based multifeature fusion sentiment analysis model (EMFSA). The model mines the sentiments of emojis, topics, and text features. Initially, a pretraining method for feature extraction is employed to enhance the semantic expressions of emotions in text by extracting contextual semantic information from emojis. Following this, a sentiment- and emoji-masked language model is designed to prioritize the masking of emojis and words with implicit sentiments, focusing on learning the emotional semantics contained in text. Additionally, we proposed a multifeature fusion method based on a cross-attention mechanism by determining the importance of each word in a text from a topic perspective. Next, this method is integrated with the original semantic information of emojis and the enhanced text features, attaining improved sentiment representation accuracy for short texts. Comparative experiments conducted with the state-of-the-art baseline methods on three public datasets demonstrate that the proposed model achieves accuracy improvements of 2.3%, 10.9%, and 2.7%, respectively, validating its effectiveness.

Cross-domain aspect-based sentiment classification with hybrid prompt

With the popularity of online shopping, an increasing number of buyers and sellers are basing their decisions on mass online reviews. As an efficient approach to obtain fine-grained sentiment information, cross-domain aspect-based sentiment classification has received extensive attention from academia and industry. However, this task encounters challenges such as limited labeled data or extreme differences in sentiment distributions between source and target domains. To address these issues, we propose a novel approach named CSC-PLDAT (Cross-domain Aspect-based Sentiment Classification with Prompt Learning and Domain Adversarial Training). Specifically, CSC-PLDAT incorporates a hybrid prompt comprising transferable and task-specific components. Based on this prompt, our model can effectively transfer domain-specific sentiment information. Meanwhile, it can adapt to the situation with limited labeled or skewed sentiment distributions. Additionally, considering various sentiment words can express identical sentiment polarities, we employ domain adversarial training to generate domain-independent sentiment representations for sentiment classification. The experiment is conducted on four popular benchmark datasets. The results demonstrate that our model outperforms the state-of-the-art methods, with an average micro-F1 score of 71.45% in cross-domain aspect-based sentiment classification. Furthermore, our model exhibits promising outcomes, particularly in scenarios characterized by limited labeled data and skewed sentiment distributions.

An autoencoder-based self-supervised learning for multimodal sentiment analysis

Representation learning is a crucial and challenging task within multimodal sentiment analysis. Effective multimodal sentiment representations contain two key aspects: consistency and difference. However, the state-of-the-art multimodal sentiment analysis approaches failed to capture the difference and consistency of sentiment information across diverse modalities. To address the multimodal sentiment representation problem, we propose an autoencoder-based self-supervised learning framework. In the pre-training stage, an autoencoder is designed for each modality, leveraging unlabeled data to learn richer sentiment representations for each modality through sample reconstruction and modality consistency detection tasks. In the fine-tuning stage, the pre-trained autoencoder is injected into MulT (AE-MT) and enhance the model's ability to extract deep sentiment information by incorporating a contrastive learning auxiliary task. Our experiments on the popular Chinese sentiment analysis benchmark (CH-SIMS v2.0) and English sentiment analysis benchmark (MOSEI) demonstrate significant gains over baseline models.

Dual-channel relative position guided attention networks for aspect-based sentiment analysis

Aspect-based sentiment analysis (ABSA) aims to match sentiment tendencies for different aspects of a sentence to understand the product experience of the user. It is a pressing challenge for existing ABSA methods to synthesize sentences’ semantic relevance and syntactic dependency for more comprehensive sentiment representations. In this paper, we propose a Dual-Channel Relative Position Guided Attention Network (Dual-RPGA). Dual-RPGA deeply learns semantic and syntactic representations of sentiment to provide reliable knowledge for dynamic fusion and prediction of sentiment. First, we design a syntactic graph attention network (Syn-GAT) to learn the syntactic relative position between aspect and context, which guides the sentiment syntactic representation. Then, we build a semantic attention network (Sem-Attention). It computes semantic attention and similarity coefficients for aspects and contexts to enhance sentiment semantic expressions. Finally, we design a fusion network (Bi-Fusion) that realizes dynamic feature interactions of sentiment semantics and syntactics to perform sentiment prediction. We conduct extensive experiments on two groups of datasets to validate the performance of Dual-RPGA on the ABSA task. The results show that Dual-RPGA outperforms the optimal baseline by 0.58%∼1.49% of the Acc score, which verifies that Dual-RPGA performs better on the ABSA task.

Learning deep news sentiment representations for macro-finance

This paper introduces custom neural network techniques to the problem of latent economic factor extraction for voluminous news analytics data. In the context of macro-financial news, we derive low-dimensional representations of time series that arise in textual sentiment analyses spanning various topics. We explore three applications for compressed news sentiment data: nowcasting GDP growth, explaining asset class returns in a panel data analysis, and time series momentum investment. Our empirical study shows that nonlinear data representations based on supervised autoencoder architectures compare favorably to alternatives across all applications. In specific, we demonstrate that augmenting autoencoders with supervision tasks based on common asset class returns and market characteristics disciplines the dimension reduction and naturally supports the transparency of resulting representations. Taken together, our findings position supervised autoencoders as attractive competitor models alongside PCA and PLS approaches.

Text-Centric Multimodal Contrastive Learning for Sentiment Analysis

Multimodal sentiment analysis aims to acquire and integrate sentimental cues from different modalities to identify the sentiment expressed in multimodal data. Despite the widespread adoption of pre-trained language models in recent years to enhance model performance, current research in multimodal sentiment analysis still faces several challenges. Firstly, although pre-trained language models have significantly elevated the density and quality of text features, the present models adhere to a balanced design strategy that lacks a concentrated focus on textual content. Secondly, prevalent feature fusion methods often hinge on spatial consistency assumptions, neglecting essential information about modality interactions and sample relationships within the feature space. In order to surmount these challenges, we propose a text-centric multimodal contrastive learning framework (TCMCL). This framework centers around text and augments text features separately from audio and visual perspectives. In order to effectively learn feature space information from different cross-modal augmented text features, we devised two contrastive learning tasks based on instance prediction and sentiment polarity; this promotes implicit multimodal fusion and obtains more abstract and stable sentiment representations. Our model demonstrates performance that surpasses the current state-of-the-art methods on both the CMU-MOSI and CMU-MOSEI datasets.

Community relations discovery methods for users in Fancircle based on sentiment analysis in China

PurposeThe identification of network user relationship in Fancircle contributes to quantifying the violence index of user text, mining the internal correlation of network behaviors among users, which provides necessary data support for the construction of knowledge graph.Design/methodology/approachA correlation identification method based on sentiment analysis (CRDM-SA) is put forward by extracting user semantic information, as well as introducing violent sentiment membership. To be specific, the topic of the implementation of topology mapping in the community can be obtained based on self-built field of violent sentiment dictionary (VSD) by extracting user text information. Afterward, the violence index of the user text is calculated to quantify the fuzzy sentiment representation between the user and the topic. Finally, the multi-granularity violence association rules mining of user text is realized by constructing violence fuzzy concept lattice.FindingsIt is helpful to reveal the internal relationship of online violence under complex network environment. In that case, the sentiment dependence of users can be characterized from a granular perspective.Originality/valueThe membership degree of violent sentiment into user relationship recognition in Fancircle community is introduced, and a text sentiment association recognition method based on VSD is proposed. By calculating the value of violent sentiment in the user text, the annotation of violent sentiment in the topic dimension of the text is achieved, and the partial order relation between fuzzy concepts of violence under the effective confidence threshold is utilized to obtain the association relation.

Do Sentence-Level Sentiment Interactions Matter? Sentiment Mixed Heterogeneous Network for Fake News Detection

With the proliferation of fake news, the spread of misleading information can easily cause social panic and group polarization. Many existing methods for detecting fake news rely on linguistic and semantic features extracted from the content of the news. Some existing approaches focus on sentiment analysis for fake news detection, but the sentiment changes and sentence-level emotional interactions in news classification are not fully analyzed. Fortunately, we observe that in long-form news, the change and mutual influence of sentiment between sentences are different. To extract the features of sentiment interaction between sentences in the article, we propose a graph attention network-based model that combines both sentiment and external knowledge comparison to meet the needs of fake news classification. We obtain the contextual sentiment representation and entity representation of the sentence through the heterogeneous network and the emotion interaction network and obtain the change of the sentiment vector through the emotion comparison network. We compare the entity vectors in the context with those corresponding knowledge base (KB)-based, combine them with the contextual semantic representation of the sentence, and finally input them into the classifier. In experiments, our model performs well in both single and multiclass classification, achieving the state-of-the-art accuracy on existing datasets.

Analyzing the online public sentiments related to Russia-Ukraine war over Twitter

Purpose Sharing and obtaining information over social media has enabled people to express their opinions regarding any event. Since the tweets regarding the Russia-Ukraine war were extensively publicized on social media, this study aims to analyse the temporal sentiments people express through tweets related to the war. Design/methodology/approach Relevant hashtag related to the Russia-Ukraine war was identified, and tweets were downloaded using Twitter API, which were later migrated to Orange Data mining software. Pre-processing techniques like transformation, tokenization, and filtering were applied to the extracted tweets. VADER (Valence Aware Dictionary for Sentiment Reasoning) sentiment analysis module of Orange software was used to categorize tweets into positive, negative and neutral ones based on the tweet polarity. For ascertaining the key and co-occurring terms and phrases in tweets and also to visualize the keyword clusters, VOSviewer, a data visualization software, was made use of. Findings An increase in the number of tweets is witnessed in the initial days, while a decline is observed over time. Most tweets are negative in nature, followed by positive and neutral ones. It is also ascertained that tweets from verified accounts are more impactful than unverified ones. russiaukrainewar, ukraine, russia, false, war, nato, zelensky and stoprussia are the dominant co-occurring keywords. Ukraine, Russia and Putin are the top hashtags for sentiment representation. India, the USA and the UK contribute the highest tweets. Originality/value The study tries to explore the public sentiments expressed over Twitter related to Russia-Ukraine war.

Learning Disentangled Representation for Multimodal Cross-Domain Sentiment Analysis.

Multimodal cross-domain sentiment analysis aims at transferring domain-invariant sentiment information across datasets to address the insufficiency of labeled data. Existing adaptation methods achieve well performance by remitting the discrepancies in characteristics of multiple modalities. However, the expressive styles of different datasets also contain domain-specific information, which hinders the adaptation performance. In this article, we propose a disentangled sentiment representation adversarial network (DiSRAN) to reduce the domain shift of expressive styles for multimodal cross-domain sentiment analysis. Specifically, we first align the multiple modalities and obtain the joint representation through a cross-modality attention layer. Then, we disentangle sentiment information from the multimodal joint representation that contains domain-specific expressive style by adversarial training. The obtained sentiment representation is domain-invariant, which can better facilitate the sentiment information transfer between different domains. Experimental results on two multimodal cross-domain sentiment analysis tasks demonstrate that the proposed method performs favorably against state-of-the-art approaches.

SenticGAT: Sentiment Knowledge Enhanced Graph Attention Network for Multi-view Feature Representation in Aspect-based Sentiment Analysis

Currently, computational intelligence methods, especially artificial neural networks, are increasingly applied to many scenarios. We mainly attempt to explore the task of fine-grained sentiment classification of review data through computational intelligence methods, especially artificial neural networks, and this task is also known as aspect-based sentiment analysis (ABSA). We propose a new technique called SenticGAT which is a multi-view features fusion model enhanced by an external sentiment database. We encode the external sentiment information into the syntactic dependency tree to obtain an enhanced graph with rich sentiment representation. Then we obtain multi-view features including semantics, syntactic, and sentiment features through GAT based on the enhanced graph by external knowledge. We also design a new strategy for fusing multi-view features using the feature parallel frame and convolution method. Eventually, the sentiment polarity of a specific aspect is determined based on the completely fused multi-view features. Experimental results on four public benchmark datasets demonstrate that our method is effective and sound. And it performs superiorly to existing approaches in fusion multiple-view features.

Learning implicit sentiments in Alzheimer's disease recognition with contextual attention features.

Alzheimer's disease (AD) is difficult to diagnose on the basis of language because of the implicit emotion of transcripts, which is defined as a supervised fuzzy implicit emotion classification at the document level. Recent neural network-based approaches have not paid attention to the implicit sentiments entailed in AD transcripts. A two-level attention mechanism is proposed to detect deep semantic information toward words and sentences, which enables it to attend to more words and fewer sentences differentially when constructing document representation. Specifically, a document vector was built by progressively aggregating important words into sentence vectors and important sentences into document vectors. Experimental results showed that our method achieved the best accuracy of 91.6% on annotated public Pitt corpora, which validates its effectiveness in learning implicit sentiment representation for our model. The proposed model can qualitatively select informative words and sentences using attention layers, and this method also provides good inspiration for AD diagnosis based on implicit sentiment transcripts.

Evaluation of Accuracy Degradation Resulting from Concept Drift in a Fake News Detection System Using Emotional Expression

Fake news on social media has become a social problem. Fake news refers to false information that is deliberately intended to deceive people. Several studies have been conducted on automatic detection systems that reduce the damage caused by fake news. However, most studies address the improvements made in detection accuracy, and real-world operations are rarely discussed. As the contents and expressions of fake news change over time, a model with a high detection accuracy loses accuracy after a few years. This phenomenon is called concept drift. As most conventional methods employ word representations, these methods exhibit accuracy degradation resulting from changes in word fads and usage. However, methods using the sentiment information of words can identify inflammatory sentences, which is a characteristic of fake news, and may suppress performance degradation caused by concept drift. In this study, a model using vector representations obtained from an emotion dictionary was compared with a model using conventional word embedding. Subsequently, we verified the resistance of the model to performance degradation. The results revealed the method using sentiment representation is less susceptible to concept drift. Models and learning methods that can achieve both detection accuracy and resistance to accuracy degradation can enable further development of fake news detection systems.

Sentiment Gradient - Improving Sentiment Analysis with Entropy Increase

Information sharing on the Web has also led to the rise and spread of fake news. Considering that fake information is generally written to trigger stronger feelings from the readers than simple facts, sentiment analysis has been widely used to detect fake news. Nevertheless, sarcasm, irony, and even jokes use similarwritten styles, making the distinction between fake and fact harder to catch automatically. We propose a new fake news Classifier that considers a set of language attributes and the gradient of sentiments contained in a message. Sentiment analysis approaches are based on labelling news with a unique value that shrinks the entire message to a single feeling. We take a broader view of a message’s sentiment representation, trying to unravel the gradient of sentiments a message may bring. We tested our approach using two datasets containing texts written in Portuguese: a public one and another we created with more up-to-date news scrapped from the Internet. Although we believe our approach is general, we tested for the Portuguese language. Our results show that the sentiment gradient positively impacts the fake news classification performance with statistical significance. The F-Measure reached 94 %, with our approach surpassing available ones (with a p-value less than 0.05 for our results).

Learning discriminative multi-relation representations for multimodal sentiment analysis

Modality representation learning is a critical issue in multimodal sentiment analysis (MSA). A good sentiment representation should contain as much effective information as possible while being discriminative enough to be better recognized. Previous attention-based MSA methods mainly rely on word-level feature interactions to capture intra-modality and inter-modality relations, which may lead to the loss of essential sentiment information. Furthermore, they primarily focus on information fusion but do not give enough importance to feature discrimination. To address these challenges, we propose a modal-utterance-temporal attention network with multimodal sentiment loss (MUTA-Net) for learning discriminative multi-relation representations, where the modal-utterance-temporal attention (MUTA) and multimodal sentiment loss (MMSL) are the two core units. First, we propose MUTA incorporate utterance-level feature vectors into the interactions of different modalities, which can help extract more useful relationships as the utterance-level vectors may contain sentiment information complementary to word-level vectors. Second, MMSL is designed to achieve a large inter-class distance and a small intra-class feature distance simultaneously in multimodal scenes to enhance the discriminative power of feature representations. Our experiments on four public multimodal datasets show that MUTA-Net outperforms previous baselines significantly.

MSFNet: modality smoothing fusion network for multimodal aspect-based sentiment analysis

Multimodal aspect-based sentiment classification (MABSC) aims to determine the sentiment polarity of a given aspect in a sentence by combining text and image information. Although the text and the corresponding image in a sample are associated with aspect information, their features are represented in distinct semantic spaces, creating a substantial semantic gap. Previous research focused primarily on identifying and fusing aspect-level sentiment expressions of different modalities while ignoring their semantic gap. To this end, we propose a novel aspect-based sentiment analysis model named modality smoothing fusion network (MSFNet). In this model, we process the unimodal aspect-aware features via the feature smoothing strategy to partially bridge modality gap. Then we fuse the smoothed features deeply using the multi-channel attention mechanism, to obtain aspect-level sentiment representation with comprehensive representing capability, thereby improving the performance of sentiment classification. Experiments on two benchmark datasets, Twitter2015 and Twitter2017, demonstrate that our model outperforms the second-best model by 1.96% and 0.19% in terms of Macro-F1, respectively. Additionally, ablation studies provide evidence supporting the efficacy of each of our proposed modules. We release the code at: https://github.com/YunjiaCai/MSFNet.

Improving aspect-based sentiment analysis with contrastive learning

As a fine-grained sentiment analysis task focusing on detecting the sentiment polarity of aspect(s) in a sentence, aspect-based sentiment analysis (ABSA) plays a significant role in opinion analysis and review analysis. Recently, a number of methods have emerged to leverage contrastive learning techniques to enhance the performance of ABSA by learning fine-grained sentiment representations. In this paper, we present and compare two commonly used contrastive learning approaches for enhancing ABSA performance: sentiment-based supervised contrastive learning and augmentation-based unsupervised contrastive learning. Sentiment-based supervised contrastive learning employs sentiment labels to distinguish between positive and negative samples. Augmentation-based unsupervised contrastive learning aims to utilize various data augmentation strategies to generate positive samples. Experimental results on three public ABSA datasets demonstrate that both contrastive learning methods significantly improve the performance of ABSA. Sentiment-based supervised contrastive learning outperforms augmentation-based unsupervised contrastive learning in terms of overall performance improvements. Furthermore, we conduct additional experiments to illustrate the effectiveness and generalizability of these two contrastive learning approaches. The experimental code and data are publicly available at the link: https://github.com/Linda230/ABSA-CL.

Learning Deep News Sentiment Representations for Macro Finance

Learning Deep News Sentiment Representations for Macro Finance

Deep Learning-based Integrated Framework for stock price movement prediction

Stock market prediction is a very important problem in the economics field. With the development of machine learning, more and more algorithms are applied in the stock market to predict the stock price movement. However, stock market prediction is regarded as a challenging task for the noise and volatility of stock market data. Therefore, in this paper, a novel hybrid model SA-DLSTM is proposed to predict stock market and simulation trading by combine a emotion enhanced convolutional neural network (ECNN), the denoising autoencoder (DAE) models, and long short-term memory model (LSTM). Firstly, user-generated comments on Internet were used as a complement to stock market data, and ECNN was applied to extract the sentiment representation. Secondly, we extract the key features of stock market data by DAE, which can improve the prediction accuracy. Thirdly, we take the timeliness of emotion on stock market into consideration and construct more reliable and realistic sentiment indexes. Finally, the key features of stock data and sentiment indexes are fed into LSTM to make stock market prediction. Experiment results show that the prediction accuracy of SA-DLSTM are superior to other compared models. Meanwhile, SA-DLSTM has a good performance both in return and risk. It can help investors make wise decisions.