Sentiment Model Research Articles

Crypto-sentiment Detection in Malay Text Using Language Models with an Attention Mechanism

Background: Due to the increased interest in cryptocurrencies, opinions on cryptocurrency-related topics are shared on news and social media. The enormous amount of sentiment data that is frequently released makes data processing and analytics on such important issues more challenging. In addition, the present sentiment models in the cryptocurrency domain are primarily focused on English with minimal work on Malay language, further complicating problems. Objective: The performance of the sentiment regression model to forecast sentiment scores for Malay news and tweets is examined in this study. Methods: Malay news headlines and tweets on Bitcoin and Ethereum are used as the input. A hybrid Generalized Autoregressive Pretraining for Language Understanding (XLNet) language model in combination with Bidirectional-Gated Recurrent Unit (Bi-GRU) deep learning model is applied in the proposed sentiment regression implementation. The effectiveness of the proposed sentiment regression model is also investigated using the multi-head self-attention mechanism. Then, a comparison analysis using Bidirectional Encoder Representations from Transformers (BERT) is carried out. Results: The experimental results demonstrate that the number of attention heads is vital in improving the XLNet-GRU sentiment model performance. There are slight improvements of 0.03 in the adjusted R2 values with an average MAE of 0.163 (Malay news) and 0.174 (Malay tweets). In addition, an average RMSE of 0.267 and 0.255 were obtained respectively for Malay news and tweets, which show that the proposed XLNet-GRU sentiment model outperforms the BERT sentiment model with lesser prediction errors. Conclusion: The proposed model contributes to predicting sentiment on cryptocurrency. Moreover, this study also introduced two carefully curated Malay corpora, CryptoSentiNews-Malay and CryptoSentiTweets-Malay, which are extracted from news and tweets, respectively. Further works to enhance Malay news and tweets corpora on cryptocurrency-related issues will be expended with implementing the proposed XLNet Bi-GRU deep learning model for greater financial insight. Keywords: Cryptocurrency, Deep learning model, Malay text, Sentiment analysis, Sentiment regression model

SENTIMENT ANALYSIS TECHNOLOGY FOR USER FEEDBACK SUPPORT IN E-COMMERCE SYSTEMS BASED ON MACHINE LEARNING

Context. The interaction between a company and its target audience has been studied for centuries. From the very beginning of commercial relations, the relationship between the service provider and the recipient has been valued almost above all else. Trade is built on trust and respect. The image of an entrepreneur is often more important than the product he sells. For hundreds of years, the relationship between the merchant and the buyer, the entrepreneur and the client has not lost its importance, and in the era of mass digitalization, the quality of the relationship between the company and the target audience of different sizes and professional feedback support with clients often start the success of e-business. To provide these additional tools and information technologies to help businessmen monitor e-business development opportunities in a specific location, as well as establish feedback with users through social networks and mass media. Obtaining such tools will significantly expand the vision of market opportunities for ebusiness, it will clarify which of them make sense to invest in, and which ones are not worth paying time for. Also see what idea has the future and what business model needs to be implemented/maintained/developed for the rapid development of territorial/interregional e-business. It will also help to understand which levers have the greatest effect for business changes: what not to touch, and what policies to change to ensure high speed in the implementation of the plan based on the analysis of relevant research results, for example, to receive: direct feedback from customers, the dynamics of changes in overall satisfaction or interest of the target audience and advantages/disadvantages from users using NLP analysis; support for the development of e-business in relation to the location of their enterprise and the best directions; – graphs of business development (improvement/deterioration) depending on the content of comments.
 Objective of the study is to develop information technology to support the development of e-business by analyzing business locations, processing feedback from users, analyzing and classifying customer feedback in real time from social networks: Twitter, Reddit, Facebook and others using deep learning and Natural methods. Language Processing of Ukrainian-speaking and Englishspeaking texts.
 Method. NLP-methods were used to analyze the opinions of users and customers. Among the methods of implementing the main functions of English-language news classification, the following machine learning methods are used: naive Bayesian classifier, logistic regression, and the method of support vectors. The Naive Bayes algorithm was used to classify Ukrainian-language user feedback, as it performs well on small amounts of data, is easy to train and operate, and works well with text data. Naive Bayes classifier is a very good option for our system and considering that the number of responses in the dataset is smaller compared to the averages.
 Results. A machine learning model was developed for the analysis and classification of Ukrainian- and English-language reviews from users of e-commerce systems.
 Conclusions. The created model shows excellent classification results on test data. The overall accuracy of the sentimental model for the analysis of Ukrainian-language content is quite satisfactory, 92.3%. The logistic regression method coped best with the task of analyzing the impact of English-language news on the financial market, which showed an accuracy of 75.67%. This is certainly not the desired result, but it is the largest indicator of all considered. The support vector method (SVM) coped somewhat worse with the task, which showed an accuracy of 72.78%, which is a slightly worse result than the one obtained thanks to the logistic regression method. And the naïve Bayesian classifier method did the worst with the task, which achieved an accuracy of 71.13%, which is less than the two previous methods.

Emotion quantification and classification using the neutrosophic approach to deep learning

Advancements in the rapidly evolving specialization of deep learning have aided in improving several natural language understanding tasks. Sentiment and emotion classification models have improved, but when it comes to fine-grained sentiment analysis, these models can perform better. Human sentiment in natural language is generally an intricate combination of emotions, which can sometimes be indeterminate, neutral, or ambiguous. In the case of fine-grained sentiment analysis, the sentiments can be very similar to each other and interconnected, e.g., anger and fear. Most deep learning systems try to solve the problem of fine-grained sentiment analysis as a classification problem. However, fine-grained sentiments might combine similar emotions with one primary emotion. Trying to solve the problem as a classification task can result in better performance on benchmarks but does not ensure a better understanding and representation of language. The proposed work explores applying neutrosophy for fine-grained sentiment analysis using large language models. Neutrosophy identifies neutralities and employs membership functions (neutral, positive, negative) to quantify an instance into Single Valued Neutrosophic Sets (SVNS). This paper introduces Refined Emotion Neutrosophic Sets (RENS) for emotions (with four emotions) and Refined Ekman’s Emotion Neutrosophic Sets (REENS) with seven emotions. In this paper, refined neutrosophic sets with membership functions are employed for each sentiment across a given taxonomy and assigned their values using the Neutrosophic Iterative Neural Clustering (NINC) algorithm proposed in this paper. It facilitates not only classifying sentiments but also quantifying the presence of each sentiment present in a given sample. It aids in better understanding and representation of samples across multiple sentiments, as in fine-grained sentiment analysis, experiments are performed on the GoEmotions dataset. The proposed approach performs on par with cross-entropy deep learning classifiers and is reproducible across different pre-trained language models.

"Erast Krutolobov's Journey to Moscow and St. Petersburg in the 30s of the XIX century" by V. Novodvorsky as a parody of the sentimental narrative model of travelogue

The subject of the study is a parody text of the famous philologist, author of the fundamental monograph on the "Letters of the Russian traveler" N. M. Karamzin, V. V. Sipovsky (pseudonym – V. Novodvorsky), little known to a wide range of readers, "Erast Krutolobov's Journey to Moscow and St. Petersburg in the 30s of the XIX century" as a parody of the narrative model of travelogue. The purpose of the study is to determine the specifics of the description of the sentimental narrative model of the travelogue in the work of V. Novodvorsky (V. V. Sipovsky), to characterize the author's reflection on the way of organizing the narrative in a parody text about the journey. The theoretical basis of the research was the works of Yu. N. Tynyanov, A. Shenle, V. M. Guminsky, O. V. Kublitskaya, Yu. V. Shatin, I. V. Banach devoted to the theory of parody and the analysis of the narrative structure of parody travelogues. To understand the specifics of V.V. Sipovsky's parody of the sentimental narrative model of travelogue, the following research methods are used: biographical, structural-typological, historical-literary. The scientific novelty of the research lies in a special material that expresses the point of view of the author-philologist, researcher of the "Karamzin canon" in the genre of travelogue, which allows us to discover the "common places" of the sentimental tradition, significantly modified in the 30s of the XIX century under the influence of new literary trends. The analysis of the ways of self-expression of the writer-literary critic in the process of creating a parody reveals research reflection not only at the level of literary play, but also through modeling the narrative structure of the text, in which not only the stages of the journey of the protagonist, Erast Krutolobov, "the event of the journey" are parodied, but also the ambivalence of ideas about the "event of telling" about the journey is expressed, contrasted sentimental narrative model and realistic.

Negativity and Positivity in the ICU: Exploratory Development of Automated Sentiment Capture in the Electronic Health Record.

To develop proof-of-concept algorithms using alternative approaches to capture provider sentiment in ICU notes. Retrospective observational cohort study. The Multiparameter Intelligent Monitoring of Intensive Care III (MIMIC-III) and the University of California, San Francisco (UCSF) deidentified notes databases. Adult (≥18 yr old) patients admitted to the ICU. We developed two sentiment models: 1) a keywords-based approach using a consensus-based clinical sentiment lexicon comprised of 72 positive and 103 negative phrases, including negations and 2) a Decoding-enhanced Bidirectional Encoder Representations from Transformers with disentangled attention-v3-based deep learning model (keywords-independent) trained on clinical sentiment labels. We applied the models to 198,944 notes across 52,997 ICU admissions in the MIMIC-III database. Analyses were replicated on an external sample of patients admitted to a UCSF ICU from 2018 to 2019. We also labeled sentiment in 1,493 note fragments and compared the predictive accuracy of our tools to three popular sentiment classifiers. Clinical sentiment terms were found in 99% of patient visits across 88% of notes. Our two sentiment tools were substantially more predictive (Spearman correlations of 0.62-0.84, p values < 0.00001) of labeled sentiment compared with general language algorithms (0.28-0.46). Our exploratory healthcare-specific sentiment models can more accurately detect positivity and negativity in clinical notes compared with general sentiment tools not designed for clinical usage.

Sentiment induced phrase-based machine translation: Robustness analysis of PBSMT with senti-module

Every type of machine translation system (i.e. neural, statistical, rule-based machine translation system) is equal important to build a sophistical hybrid machine translation system. Keeping this fact in my mind, I concentrate to improve statistical machine translation system with more natural way. In this paper, I try to preserve sentiment after translation to improve the overall accuracy of the machine translation system. So, I introduced senti-model here. A senti-model (sentiment model), translation model, language model, and distortion model are incorporated on the top of the beam search algorithm for decoding. At first, sentiment information is learned and modeled with translation probability by using this algorithm. Thereafter, I decode the source sentences-based on the contextual information. Overall procedure of translation modeling with a sentiment, parameter estimation for it, and senti-translation decoding (decoding with the sentiment model) are presented with empirical evidence. Experiments on a benchmark English–Hindi dataset shows that the proposed model is capable to improve the accuracy (in terms of 4.66 BLEU points, 4.09 LeBleu points, 4.67 NIST points, 5.71 RIBES points) significantly and preserves sentiment 7.79% more than the state-of-the-art technique.

Sentiment Analysis on Large-Scale Covid-19 Tweets using Hybrid Convolutional LSTM Based on Naïve Bayes Sentiment Modeling

Millions of lives were affected rapidly throughout the world when the Covid-19 outbreak spread by leaps and bounds. During this catastrophic period, people used to express their condolence as well as emotions through different social networks. In order to analyze the public comments on Twitter, an experimental approach is developed based on popular words regarding this pandemic. In this paper, various NLP-based research works are discussed on sentiment analysis, trend prediction, topic modeling, learning mechanisms, etc. Furthermore, the hybrid deep learning models are developed based on the Naïve Bayes sentiment model to predict the sentiment from the collected huge number of Coronavirus-related tweets. After performing the n-gram analysis, the Covid-19 specific words are extracted based on their popularity. The public sentiment trend has been analyzed using the extracted topics related to Covid-19 and the tweets are classified according to their sentiment scores. The distinguished sentiment ratings are assigned to the collected tweets based on their sentiment class. Then Convo-Sequential and Convo-Bidirectional long-short term networks are trained using fine-grained sentiment-rated tweets to categorize Covid-19 tweets into five different sentiment classes. Finally, our proposed Convo-Sequential and Convo-Bidirectional LSTM models achieved 84.52% and 85.03% of validation accuracy respectively for the first phase dataset whereas using the second phase dataset the models obtained the validation accuracy of 86.58% and 87.22% respectively.

Tourists’ Perception and Influence Factors in Virtual Tourism Using Bayesian Sentimental Analysis Model in Vietnam Based on e WOM for Sustainable Development

Objective: The advancement of Internet technology brought up the tourism industry towards new development and opportunities. With application of the Internet technology tourism industry comprises a vast range of virtual communities such as Trip Advisor, Agoda, Travelocity and so on. Existing research concentrated on evaluating the factors influencing virtual communities’ behaviour with limited evaluation of tourist perception. This paper focused on examining the tourists' perception of a virtual tour through the sentimental analysis model based on eWOM for sustainable development. Method: The developed model comprises the group average Bayesian network with the computation of the polarity of the tourist perception. A Bayesian network is a data-driven method involved in estimating dependence among the variable with probabilistic computation. Results: The analysis is based on data collected from sample population in Vietnam with consideration of the 11 variables. Participation intensity, social identity, functional value, emotional value, sharing, interaction, and user satisfaction are among eleven primary variables that have been chosen. Conclusions: The analysis of the results expressed that the user satisfaction level is based on the user's experience and functional value. Additionally, the analysis stated that social value comprises the intermediary role in virtual tourism. This research adds to research methodologies of user engagement methods as well as serves as a reference for theoretical research and management practise in the virtual tourist community.

Sentiment and Topic Modeling Analysis on Twitter Reveals Concerns over Cannabis-Containing Food after Cannabis Legalization in Thailand.

Twitter has been used to express a diverse range of public opinions about cannabis legalization in Thailand. The purpose of this study was to observe changes in sentiments after cannabis legalization and to investigate health-related topics discussed on Twitter. Tweets in Thai and English related to cannabis were scraped from Twitter between May 1 and June 13, 2022, during cannabis legalization in Thailand. Sentiment and topic-modeling analyses were used to compare the content of tweets before and after legalization. Health-related topics were manually grouped into categories by their content and rated according to the number of corresponding tweets. We collected 21,242 and 6,493 tweets, respectively, for Thai and English search terms. A sharp increase in the number of tweets related to cannabis legalization was detected at the time of its public announcement. Sentiment analysis in the Thai search group showed a significant change (p < 0.0001) in sentiment distribution after legalization, with increased negative and decreased positive sentiments. A significant change was not found in the English search group (p = 0.4437). Regarding cannabis-containing food as a leading issue, topic-modeling analysis revealed public concerns after legalization in the Thai search group, but not the English one. Topics related to cannabis tourism surfaced only in the English search group. Since cannabis legalization, the primary health-related concern has been cannabis-containing food. Education and clear regulations on cannabis use are required to strengthen oversight of cannabis in the Thai population, as well as among medical tourists.

Enhancing Collaborative Filtering-Based Recommender System Using Sentiment Analysis

Recommendation systems (RSs) are widely used in e-commerce to improve conversion rates by aligning product offerings with customer preferences and interests. While traditional RSs rely solely on numerical ratings to generate recommendations, these ratings alone may not be sufficient to offer personalized and accurate suggestions. To overcome this limitation, additional sources of information, such as reviews, can be utilized. However, analyzing and understanding the information contained within reviews, which are often unstructured data, is a challenging task. To address this issue, sentiment analysis (SA) has attracted considerable attention as a tool to better comprehend a user’s opinions, emotions, and attitudes. In this study, we propose a novel RS that leverages ensemble learning by integrating sentiment analysis of textual data with collaborative filtering techniques to provide users with more precise and individualized recommendations. Our system was developed in three main steps. Firstly, we used unsupervised “GloVe” vectorization for better classification performance and built a sentiment model based on Bidirectional Long Short-Term Memory (Bi-LSTM). Secondly, we developed a recommendation model based on collaborative filtering techniques. Lastly, we integrated our sentiment analysis model into the RS. Our proposed model of SA achieved an accuracy score of 93%, which is superior to other models. The results of our study indicate that our approach enhances the accuracy of the recommendation system. Overall, our proposed system offers customers a more reliable and personalized recommendation service in e-commerce.

Natural Language Processing used in sentiment analysis of poetry: a study of six common techniques

This paper explores and compares the accuracies of six different NLP model techniques when applied to analysis of English poetry. It was found that bigram-based and word embedding-based models were the most accurate in deriving emotions from the bodies of text in the corpus, with respective accuracies of 61.67% and 61.68%. The least accurate models were the unigram-based model and distributed dictionary-based model, constructed using the traditional approach rather than the new methodology, and had accuracies of 11.173% and 48.17% respectively. Most models passed the benchmark accuracies of 49.42% and 47.86%, the higher accuracy one being a word count model and lower one being a sentiment model. The need for newer methodologies that allow for higher dimensional levels of semantic analysis to be performed is also discussed in this paper, along with the potential impacts of this research to the field of Natural Language Processing.

Arachne: Search-Based Repair of Deep Neural Networks

The rapid and widespread adoption of Deep Neural Networks (DNNs) has called for ways to test their behaviour, and many testing approaches have successfully revealed misbehaviour of DNNs. However, it is relatively unclear what one can do to correct such behaviour after revelation, as retraining involves costly data collection and does not guarantee to fix the underlying issue. This article introduces Arachne, a novel program repair technique for DNNs, which directly repairs DNNs using their input-output pairs as a specification. Arachne localises neural weights on which it can generate effective patches and uses differential evolution to optimise the localised weights and correct the misbehaviour. An empirical study using different benchmarks shows that Arachne can fix specific misclassifications of a DNN without reducing general accuracy significantly. On average, patches generated by Arachne generalise to 61.3% of unseen misbehaviour, whereas those by a state-of-the-art DNN repair technique generalise only to 10.2% and sometimes to none while taking tens of times more than Arachne. We also show that Arachne can address fairness issues by debiasing a gender classification model. Finally, we successfully apply Arachne to a text sentiment model to show that it generalises beyond convolutional neural networks.

Multimodal knowledge graph construction of Chinese traditional operas and sentiment and genre recognition

The advancement of digital technologies promotes the documentation of traditional operas, leaving a large amount of data but in a state of fragmentation. Constructing a knowledge graph (KG) is an effective way to realize the knowledge integration and reduce fragmentation, which can help the public understand traditional operas. However, constructed KGs in cultural heritage are mainly unimodal, lacking the ability to give the public a comprehensive perception, especially when they do not have related in-depth knowledge. In this paper, we take Chinese nation-level traditional operas as an example and construct a traditional opera ontology OpeOnto including classes with deep semantics (topic, sentiment). Then, we adopt an OpeOnto-driven way to construct multimodal knowledge graph OpeMKG including images and music links from several data resources. Aimed at analysing sentiments in OpeMKG and the automatic genre recognition of works for the preparation of automatic updating of OpeMKG, we develop a novel unified sentiment and genre recognition model (SGRM) for traditional operas with multimodal fusion and multi-task learning (MTL). The proposed model is examined on the built dataset of traditional operas and experimental results demonstrate its superiority compared with several state-of-the-art baselines.

Signatures of the uncanny valley effect in an artificial neural network

Robots and computer graphics characters that resemble humans but are not perfectly human-like tend to evoke negative feelings in human observers, which is known as the “uncanny valley effect.” In this study, we used a recent artificial neural network called Contrastive Language-Image Pre-training (CLIP) that learns visual concepts from natural language supervision as a visual sentiment model for humans to examine the semantic match between images with graded manipulation of human-likeness and words used in previous studies to describe the uncanny valley effect. Our results showed that CLIP estimated the matching of words of negative valence to be maximal at the midpoint of the transition from a human face to other objects, thereby indicating the signature of the uncanny valley effect. Our findings suggest that visual features characteristic to the conflicts of visual cues, particularly cues related to human faces, are associated with negative verbal expressions in our everyday experiences, and CLIP learned such an association from the training datasets. Our study is a step toward exploring how visual cues are related to human observers’ sentiment using a novel psychological platform, that is, an artificial neural network.

From Twitter to Aso-Rock: A sentiment analysis framework for understanding Nigeria 2023 presidential election

IntroductionSocial media platforms such as Facebook, LinkedIn, Twitter, among others have been used as tools for staging protests, opinion polls, campaign strategy, medium of agitation and a place of interest expression especially during elections. AimIn this work, a Natural Language Processing framework is designed to understand Nigeria 2023 presidential election based on public opinion using Twitter dataset. MethodsTwo million tweets with 18 features were collected from Twitter containing public and personal tweets of the three top contestants – Atiku Abubakar, Peter Obi and Bola Tinubu – in the forthcoming 2023 Presidential election. Sentiment analysis was performed on the preprocessed dataset using three machine learning models namely: Long Short-Term Memory (LSTM) Recurrent Neural Network, Bidirectional Encoder Representations from Transformers (BERT) and Linear Support Vector Classifier (LSVC) models. This study spanned ten weeks starting from the candidates’ declaration of intent to run for Presidency. ResultsThe sentiment models gave an accuracy, precision, recall, AUC and f-measure of 88%, 82.7%, 87.2%, 87.6% and 82.9% respectively for LSTM; 94%, 88.5%, 92.5%, 94.7% and 91.7% respectively for BERT and 73%, 81.4%, 76.4%, 81.2% and 79.2% respectively for LSVC. Result also showed that Peter Obi has the highest total impressions the highest positive sentiments, Tinubu has the highest network of active friends while Atiku has the highest number of followers. ConclusionSentiment analysis and other Natural Language Understanding tasks can aid in the understanding of the social media space in terms of public opinion mining. We conclude that opinion mining from Twitter can form a general basis for generating insights for election as well as modeling election outcomes.

Neural Network and Sentimental Model for Prediction of Stock Trade Value

Neural Network and Sentimental Model for Prediction of Stock Trade Value

Negative Sentiment Modeling and Public Legal Liability from Urban Green Space: A Framework for Policy Action in China

This paper intends to optimize the urban green space (UGS) management and implementation strategies by analyzing climate change models and reviewing economic, energy, and public health policies. This paper studies the public perception of climate change-induced public health emergency (PHE) in China by surveying online public comments. Specifically, it looks into public health perception, anxiety perception, relative deprivation, and emotional polarity from public online comments. The following conclusions are drawn through the empirical test of 179 questionnaires. The findings revealed that health risk perception has a positive predictive effect on relative deprivation and anxiety perception. The higher the health risk perception, the stronger the relative deprivation and anxiety are. Anxiety perception and relative deprivation have mediating effects in the model. In addition, the main research method adopts a questionnaire survey. The mediating effect between each variable is further studied. This paper analyzes the citizens’ right to health and public health protection under climate change, and explains public risk perception and anxiety perception. Meanwhile, the evaluation cases are used to analyze the public health and UGS construction strategies to suggest climate compensation laws and improve the urban greening rate. This finding has practical reference value for promoting the deep integration of UGS and public health. It can promote the development and planning of UGS under climate change and biodiversity loss and has significant reference value for improving negative emotions and the public legal liability system.

Improved Feature Representation Using Collaborative Network for Cross-Domain Sentiment Analysis

Sentiment Analysis task helps us to estimate the opinion of a person from his reviews or comments about a product, person, politics, etc., Cross-Domain Sentiment Analysis (CDSA) empowers the Sentiment models with the ability to forecast the opinion of a review coming from a different domain other than the domain where the model is trained. The challenge of the CDSA model relies on bridging the relationship between words in the source and target domain. Several types of research in CDSA focus on determining the domain invariant features to adapt the model to the target domain, such model shows less focus on aspect terms of the sentence. We propose CWAN (Collaborative Word Attention Network), which integrates aspects and domain invariant features of the sentences to calculate the sentiment. CWAN uses attention networks to capture the domain-independent features and aspects of the sentences. The sentence and aspect attention models are executed collaboratively to determine the sentiment of the sentence. Amazon product review dataset is used in this experiment. The performance of the CWAN model is compared with other baseline CDSA models. The results show that CWAN outperforms other baseline models.

PyFin-sentiment: Towards a machine-learning-based model for deriving sentiment from financial tweets

Responding to the poor performance of generic automated sentiment analysis solutions on domain-specific texts, we collect a dataset of 10,000 tweets discussing the topics of finance and investing. We manually assign each tweet its market sentiment, i.e., the investor’s anticipation of a stock’s future return. Using this data, we show that all existing sentiment models trained on adjacent domains struggle with accurate market sentiment analysis due to the task’s specialized vocabulary. Consequently, we design, train, and deploy our own sentiment model. It outperforms all previous models (VADER, NTUSD-Fin, FinBERT, TwitterRoBERTa) when evaluated on Twitter posts. On posts from a different platform, our model performs on par with BERT-based large language models. We achieve this result at a fraction of the training and inference costs due to the model’s simple design. We publish the artifact as a python library to facilitate its use by future researchers and practitioners.

Integrating Economic Theory, Domain Knowledge, and Social Knowledge into Hybrid Sentiment Models for Predicting Crude Oil Markets

For several decades, sentiment analysis has been considered a key indicator for assessing market mood and predicting future price changes. Accurately predicting commodity markets requires an understanding of fundamental market dynamics such as the interplay between supply and demand, which are not considered in standard affective models. This paper introduces two domain-specific affective models, CrudeBERT and CrudeBERT+, that adapt sentiment analysis to the crude oil market by incorporating economic theory with common knowledge of the mentioned entities and social knowledge extracted from Google Trends. To evaluate the predictive capabilities of these models, comprehensive experiments were conducted using dynamic time warping to identify the model that best approximates WTI crude oil futures price movements. The evaluation included news headlines and crude oil prices between January 2012 and April 2021. The results show that CrudeBERT+ outperformed RavenPack, BERT, FinBERT, and early CrudeBERT models during the 9-year evaluation period and within most of the individual years that were analyzed. The success of the introduced domain-specific affective models demonstrates the potential of integrating economic theory with sentiment analysis and external knowledge sources to improve the predictive power of financial sentiment analysis models. The experiments also confirm that CrudeBERT+ has the potential to provide valuable insights for decision-making in the crude oil market.