Domain-Specific Sentiment Analysis: An Optimized Deep Learning Approach for the Financial Markets
Open accessThe DOI matches its Crossref record; 32 of the 32 references it cites were found in a registry or an index.
Abstract
Although different studies are caried out by deep learning models for financial markets sentiment analysis, there is a lack of specific embedding method that regards the domain. Therefore, the goal of this study is to discover what type of embedding techniques along with different classification algorithms work better for the financial markets’ sentiment analysis to present an optimized embedding method in the domain. In this paper we present a broad comparative study of multiple classification models trained for sentiment analysis and will improve their performance with an optimized embedding layer. We use a heterogeneous corpus of both formal (news headlines) and informal (tweets) text to increase the robustness and build the models with CBOW, GloVe, and BERT pre-trained embeddings as well as developing an optimized embedding layer to improve the results. The best results reported here are by our LSTM model with the fine-tuned embedding layer, which has an accuracy of 0.84 and a macro-average F1-score of 0.8. Our results give evidence that the fine-tuned embedding is superior to utilising pretrained CBOW, GloVe, and BERT embeddings for financial markets sentiment analysis. We train SVM, MLP, CNN, generic RNN and LSTM models by a comprehensive approach in input data and algorithms. As a result, a sentiment analysis model is presented with a robust performance for different datasets in the domain.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173 · cited 2×
“• BERT embeddings [10]: BERT stands for bidirectional encoder representation from transformers, which is a new pre-trained language representation model.”
BackgroundMethodBERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1423 · Cited by 54,325 · cited 2×
“• BERT embeddings [10]: BERT stands for bidirectional encoder representation from transformers, which is a new pre-trained language representation model.”
BackgroundMethodGlove: Global Vectors for Word Representation
Pennington, J., Socher, R., & Manning, C. (2014) · Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.3115/v1/d14-1162 · Cited by 34,071 · cited 2×
“( [6], the words in a document are represented regarding their context.”
BackgroundEfficient Estimation of Word Representations in Vector Space
Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013) · arXiv · doi:10.48550/arxiv.1301.3781 · Cited by 26,568 · cited 3×
“• CBOW embeddings [5]: A pre-trained embedding algorithm based on neural networks.”
MethodInfluentialThumbs up?: sentiment classification using machine learning techniques
Pang, B., Lee, L., & Vaithyanathan, S. (2002) · Proceedings of the ACL-02 conference on Empirical methods in natural language processing - EMNLP '02 · doi:10.3115/1118693.1118704 · Cited by 7,024 · cited 1×
“While in lexicon-based models the sentiment of a text document depends on the labels of words in a lexicon [19], in ML-based models, the sentiment is classified by a ML model trained on a labelled dataset [20].”
BackgroundThe use of ranks to avoid the assumption of normality implicit in the analysis of variance
Friedman, M. (1937) · Journal of the American Statistical Association · doi:10.1080/01621459.1937.10503522 · Cited by 7,005 · cited 2×
“One common pre-hoc test is Friedman test [27], which is a non-parametric test to assess significant differences between the results. doing Friedman test, the associated p-value with the test statistics can be determined.”
MethodInfluentialMultiple Comparisons Using Rank Sums
Dunn, O. J. (1964) · Technometrics · doi:10.1080/00401706.1964.10490181 · Cited by 5,085 · cited 1×
“There are different post-hoc tests based on rank sums such as Dunn [28], Nemenyi [29], and Games-Howell [30] that can be used.”
MethodIntroduction to WordNet: An On-line Lexical Database*
Miller, G. A., Beckwith, R., Fellbaum, C., Gross, D., & Miller, K. J. (1990) · International Journal of Lexicography · doi:10.1093/ijl/3.4.235 · Cited by 4,197 · cited 1×
“to the WordNet dictionary [4], the word ‘‘strong’’ has a positive subjectivity.”
Lexicon-based methods for sentiment analysis
Taboada, M., Brooke, J., Tofiloski, M., Voll, K., & Stede, M. (2011) · Computational Linguistics · doi:10.1162/coli_a_00049 · Cited by 3,301
Thumbs up or thumbs down?: semantic orientation applied to unsupervised classification of reviews
Turney, P. D. (2001) · Proceedings of the 40th Annual Meeting on Association for Computational Linguistics - ACL '02 · doi:10.3115/1073083.1073153 · Cited by 2,352 · cited 1×
“While in lexicon-based models the sentiment of a text document depends on the labels of words in a lexicon [19], in ML-based models, the sentiment is classified by a ML model trained on a labelled dataset [20].”
BackgroundDetermining the sentiment of opinions
Kim, S.-M., & Hovy, E. (2004) · Proceedings of the 20th international conference on Computational Linguistics - COLING '04 · doi:10.3115/1220355.1220555 · Cited by 1,482 · cited 1×
“We also expand the lexicon with the synonyms of both positive and negative labelled words by an iterative approach based on WordNet synsets [22].”
MethodMultiple Comparisons Using Rank Sums
Dunn, O. J. (1964) · Technometrics · doi:10.2307/1266041 · Cited by 865
FinBERT: Financial Sentiment Analysis with Pre-trained Language Models
Araci, D. (2019) · arXiv · Cited by 602 · cited 1×
“In another study [14], a derivative BERT embedding called Fin-BERT is pre-trained on the financial corpus.”
MethodSentiment Analysis Based on Deep Learning: A Comparative Study
Dang, N. C., Moreno-García, M. N., & De la Prieta, F. (2020) · Electronics · doi:10.3390/electronics9030483 · Cited by 593 · cited 2×
“[11], the most popular deep learning algorithms including MLP, CNN and RNN are investigated and compared to each other on multiple datasets.”
MethodGood debt or bad debt: Detecting semantic orientations in economic texts
Malo, P., Sinha, A., Korhonen, P., Wallenius, J., & Takala, P. (2014) · Journal of the Association for Information Science and Technology · doi:10.1002/asi.23062 · Cited by 513 · cited 1×
“The second one [17] consists of 4,821 various news headlines about financial markets.”
BackgroundBarber, B. M., & Odean, T. (2001) · Journal of Economic Perspectives · doi:10.1257/jep.15.1.41 · Cited by 461
Sentiment analysis on product reviews based on weighted word embeddings and deep neural networks
Onan, A. (2021) · Concurrency and Computation: Practice and Experience · doi:10.1002/cpe.5909 · Cited by 396 · cited 1×
“In the deep learning experiments, the embedding method can have a big impact on the accuracy of the model [25].”
MethodBig Data: Deep Learning for financial sentiment analysis
Sohangir, S., Wang, D., Pomeranets, A., & Khoshgoftaar, T. M. (2018) · Journal of Big Data · doi:10.1186/s40537-017-0111-6 · Cited by 390 · cited 2×
“Because of their structure and ability to extract best features, deep learning models have achieved promising results [9].”
BackgroundPairwise Multiple Comparison Procedures with Unequal N's and/or Variances: A Monte Carlo Study
Games, P. A., & Howell, J. F. (1976) · Journal of Educational Statistics · doi:10.2307/1164979 · Cited by 361
Sentiment Embeddings with Applications to Sentiment Analysis
Tang, D., Wei, F., Qin, B., Yang, N., Liu, T., & Zhou, M. (2016) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2015.2489653 · Cited by 335 · cited 1×
“The pre-trained vector representations of such words cannot be used reliably to compare their similarity and classify them sentimentally [23].”
BackgroundDistinguishing between facts and opinions for sentiment analysis: Survey and challenges
Chaturvedi, I., Cambria, E., Welsch, R. E., & Herrera, F. (2018) · Information Fusion · doi:10.1016/j.inffus.2017.12.006 · Cited by 292 · cited 1×
“While typically SA involves classifying sentiments in three categories (positive, negative, neutral), the neutral label has been disregarded in some research works [3], [13], which in turn limits the study [15].”
BackgroundAssessment Metrics for Imbalanced Learning
Japkowicz, N. (2013) · Imbalanced Learning · doi:10.1002/9781118646106.ch8 · Cited by 160 · cited 1×
“The model Furthermore, a proper metric for classifiers on imbalanced data is receiver operating characteristic or ROC curve which is not biased towards majority or minority classes [26].”
BackgroundEnhanced news sentiment analysis using deep learning methods
Souma, W., Vodenska, I., & Aoyama, H. (2019) · Journal of Computational Social Science · doi:10.1007/s42001-019-00035-x · Cited by 146 · cited 3×
“[13] carry out SA by deep learning with the aim of predicting market movements.”
BackgroundImproving Twitter Sentiment Classification Using Topic-Enriched Multi-Prototype Word Embeddings
Ren, Y., Zhang, Y., Zhang, M., & Ji, D. (2016) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v30i1.10370 · Cited by 126 · cited 2×
“This embedding issue is due to the fact that most pre-trained embedding methods regard the general application of a word in a context, while disregarding the topic information of the specific document, in which the word is used [7].”
BackgroundGómez-de-Mariscal, E., Guerrero, V., Sneider, A., Jayatilaka, H., Phillip, J. M., Wirtz, D., & Muñoz-Barrutia, A. (2021) · Scientific Reports · doi:10.1038/s41598-021-00199-5 · Cited by 104
Aspect-Based Financial Sentiment Analysis using Deep Learning
Jangid, H., Singhal, S., Shah, R. R., & Zimmermann, R. (2018) · Companion of the The Web Conference 2018 on The Web Conference 2018 - WWW '18 · doi:10.1145/3184558.3191827 · Cited by 82 · cited 1×
“In a study by Jangid et al. [12], deep learning models are applied with the aim of polarity prediction as well as aspect extraction.”
Financial Sentiment Analysis(FSA): A Survey
Man, X., Tong, L., & Lin, J. (2019) · IEEE International Conference on Industrial Cyber-Physical Systems (ICPS) · doi:10.1109/icphys.2019.8780312 · Cited by 72 · cited 2×
“The predominant deep learning algorithms which have been used for financial markets’ SA include convolutional neural networks (CNN), recurrent neural networks (RNN), and long short-term memory (LSTM) networks [2].”
BackgroundDomain-specific sentiment analysis using contextual feature generation
Choi, Y., Kim, Y., & Myaeng, S.-H. (2009) · Proceedings of the 1st international CIKM workshop on Topic-sentiment analysis for mass opinion · doi:10.1145/1651461.1651469 · Cited by 64 · cited 1×
“Thus, a successful model could handle SA challenges at sentence level, which are primarily context sensitivities and the dependencies that words have on each other in a sequence as a sentence [24].”
BackgroundFinancial markets sentiment analysis: developing a specialized Lexicon
Yekrangi, M., & Abdolvand, N. (2020) · Journal of Intelligent Information Systems · doi:10.1007/s10844-020-00630-9 · Cited by 49 · cited 1×
“In order to build the pre-defined lexicon, we used the financial market’s sentiment lexicon [21], developed in our previous work as the seed words resource to be expanded.”
MethodSelf-citationAnalysis of news sentiments using natural language processing and deep learning
Vicari, M., & Gáspari, M. (2020) · AI & Society · doi:10.1007/s00146-020-01111-x · Cited by 46 · cited 4×
“In a similar work by Vicari and Gaspari [8], LSTM is applied to the embedding layer’s vectors.”
BackgroundMethodInfluentialA Comparison of Neural Network Methods for Accurate Sentiment Analysis of Stock Market Tweets
Tabari, N., Seyeditabari, A., Peddi, T., Hadžikadić, M., & Zadrozny, W. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-13463-1_4 · Cited by 20 · cited 1×
“The first one [16] consists of 10,631 tweets about stock markets.”
BackgroundDeep Learning Approaches for Sentiment Analysis on Financial Microblog Dataset
Yıldırım, S., Jothimani, D., Kavaklioğlu, C., & Bener, A. (2019) · IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata47090.2019.9006056 · Cited by 17 · cited 4×
“[3], both traditional machine learning and deep learning algorithms are applied on another dataset of tweets collected from StockTwits and the results are compared.”
BackgroundMethodInfluential
Research works printed but not linked: 2
- 10.2307/2279372 · doi:10.2307/2279372 · Paper page
- nemenyi, 1963, Distribution-free multiple comparisons
Cited by
Recent advancements and challenges of NLP-based sentiment analysis: A state-of-the-art review
Jim, J. R., Talukder, M. A. R., Malakar, P., Kabir, M. M., Nur, K., & Mridha, M. F. (2024) · Natural Language Processing Journal · doi:10.1016/j.nlp.2024.100059 · Cited by 297
Comprehensive review and comparative analysis of transformer models in sentiment analysis
Bashiri, H., & Naderi, H. (2024) · Knowledge and Information Systems · doi:10.1007/s10115-024-02214-3 · Cited by 117
On Assessing the Performance of LLMs for Target-Level Sentiment Analysis in Financial News Headlines
Muhammad, I., & Rospocher, M. (2025) · Algorithms · doi:10.3390/a18010046 · Cited by 14
Verma, B., Meel, P., & Vishwakarma, D. K. (2025) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2025.111195 · Cited by 12
Nath, D., & Dwivedi, S. K. (2024) · International Conference on Signal Processing and Advance Research in Computing (SPARC) · doi:10.1109/sparc61891.2024.10828959 · Cited by 3
Metadata
- Authors
- Yekrangi, M., & Nikolov, N. S.
- Affiliations
- University of Limerick (IE)
- Venue
- IEEE Access
- Published
- 2023
- Volume
- 11
- Pages
- 70248-70262
- ISSN
- 2169-3536 (ISSN-L)
- OpenAlex
- W4383751614
- Cited by
- 14
- References
- 30
- Citations by year
- 2026: 3 · 2025: 6 · 2024: 4 · 2023: 1
- FWCI
- 1.24
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Locations
- IEEE Access (publishedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Figshare (submittedVersion, cc-by-nc-sa)
- University of Limerick Institutional Repository (University of Limerick) (submittedVersion, cc-by-nc-sa)
- University of Limerick Institutional Repository (University of Limerick)
- Funders
- Science Foundation Ireland (18/CRT/6223)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Sentiment Analysis and Opinion Mining
- Social Sciences › Decision Sciences › Management Science and Operations Research › Stock Market Forecasting Methods
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Keywords
- financial sentiment analysis, domain-specific embeddings, financial news headlines, tweets, CBOW, gloves, BERT, LSTM, text classification, classifier comparison
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.