A Deep Learning-Based Approach to Constructing a Domain Sentiment Lexicon: a Case Study in Financial Distress Prediction

Incremental learning algorithm for dynamic evolution of domain specific vocabulary with its stability and plasticity analysis

Domain-specific vocabulary, which is crucial in fields such as Information Retrieval and Natural Language Processing, requires continuous updates to remain effective. Incremental Learning, unlike conventional methods, updates existing knowledge without retraining from scratch. This paper presents an incremental learning algorithm for updating domain-specific vocabularies. It introduces DocLib, an archive used to capture a compact footprint of previously seen data and vocabulary terms. Task-based evaluation measures the effectiveness of the updated vocabulary by using vocabulary terms to perform a downstream task of text classification. The classification accuracy gauges the effectiveness of the vocabulary in discerning unseen documents related to the domain. Experiments illustrate that multiple incremental updates maintain vocabulary relevance without compromising its effectiveness. The proposed algorithm ensures bounded memory and processing requirements, distinguishing it from conventional approaches. Novel algorithms are introduced to assess the stability and plasticity of the proposed approach, demonstrating its ability to assimilate new knowledge while retaining old insights. The generalizability of the vocabulary is tested across datasets, achieving 97.89% accuracy in identifying domain-related data. A comparison with state-of-the-art techniques using a benchmark dataset confirms the effectiveness of the proposed approach. Importantly, this approach extends beyond classification tasks, potentially benefiting other research fields.

Construction of an Emotional Lexicon of Patients With Breast Cancer: Development and Sentiment Analysis

BACKGROUND

The innovative method of sentiment analysis based on an emotional lexicon shows prominent advantages in capturing emotional information, such as individual attitudes, experiences, and needs, which provides a new perspective and method for emotion recognition and management for patients with breast cancer (BC). However, at present, sentiment analysis in the field of BC is limited, and there is no emotional lexicon for this field. Therefore, it is necessary to construct an emotional lexicon that conforms to the characteristics of patients with BC so as to provide a new tool for accurate identification and analysis of the patients' emotions and a new method for their personalized emotion management.

OBJECTIVE

This study aimed to construct an emotional lexicon of patients with BC.

METHODS

Emotional words were obtained by merging the words in 2 general sentiment lexicons, the Chinese Linguistic Inquiry and Word Count (C-LIWC) and HowNet, and the words in text corpora acquired from patients with BC via Weibo, semistructured interviews, and expressive writing. The lexicon was constructed using manual annotation and classification under the guidance of Russell's valence-arousal space. Ekman's basic emotional categories, Lazarus' cognitive appraisal theory of emotion, and a qualitative text analysis based on the text corpora of patients with BC were combined to determine the fine-grained emotional categories of the lexicon we constructed. Precision, recall, and the F1-score were used to evaluate the lexicon's performance.

RESULTS

The text corpora collected from patients in different stages of BC included 150 written materials, 17 interviews, and 6689 original posts and comments from Weibo, with a total of 1,923,593 Chinese characters. The emotional lexicon of patients with BC contained 9357 words and covered 8 fine-grained emotional categories: joy, anger, sadness, fear, disgust, surprise, somatic symptoms, and BC terminology. Experimental results showed that precision, recall, and the F1-score of positive emotional words were 98.42%, 99.73%, and 99.07%, respectively, and those of negative emotional words were 99.73%, 98.38%, and 99.05%, respectively, which all significantly outperformed the C-LIWC and HowNet.

CONCLUSIONS

The emotional lexicon with fine-grained emotional categories conforms to the characteristics of patients with BC. Its performance related to identifying and classifying domain-specific emotional words in BC is better compared to the C-LIWC and HowNet. This lexicon not only provides a new tool for sentiment analysis in the field of BC but also provides a new perspective for recognizing the specific emotional state and needs of patients with BC and formulating tailored emotional management plans.

Examining financial distress of the Vietnamese listed firms using accounting-based models

Financial distress is generally considered the most severe consequence for firms with poor financial performance. The emergence of the Covid-19 pandemic has adversely impacted the global business system and exacerbated the number of financially distressed firms in many countries. Only firms with strong financial fundamentals can survive extreme events such as the Covid-19 pandemic and the ongoing Russia-Ukraine conflict. Vietnam is no exception. However, studies examining financial distress using accounting-based indicators, particularly at the industry level, have largely been ignored in the Vietnamese context, particularly with the emergence of the Covid-19 pandemic. This study, therefore, comprehensively examines financial distress for 500 Vietnamese listed firms during the 2012-2021 period. Our study uses interest coverage and times-interest-earned ratios to proxy a firm's financial distress. First, our findings confirm the validity of Altman's Z"- score model in Vietnam only when the interest coverage ratio is used as a proxy for financial distress. Second, our empirical findings indicate that only four financial ratios, including EBIT/Total Assets, Net Income/Total Assets, Total Liabilities/Total Assets, and Total Equity/Total Liabilities, can be used in predicting financial distress in Vietnam. Third, our analysis at the industry level indicates that the "Construction & Real Estates" industry, a significant contributor to the national economy, exhibits the most significant risk exposure, particularly during the Covid-19 pandemic. Policy implications have emerged based on the findings from this study.

Speech emotion recognition and text sentiment analysis for financial distress prediction

In recent years, there has been an increasing interest in text sentiment analysis and speech emotion recognition in finance due to their potential to capture the intentions and opinions of corporate stakeholders, such as managers and investors. A considerable performance improvement in forecasting company financial performance was achieved by taking textual sentiment into account. However, far too little attention has been paid to managerial emotional states and their potential contribution to financial distress prediction. This study seeks to address this problem by proposing a deep learning architecture that uniquely combines managerial emotional states extracted using speech emotion recognition with FinBERT-based sentiment analysis of earnings conference call transcripts. Thus, the obtained information is fused with traditional financial indicators to achieve a more accurate prediction of financial distress. The proposed model is validated using 1278 earnings conference calls of the 40 largest US companies. The findings of this study provide evidence on the essential role of managerial emotions in predicting financial distress, even when compared with sentiment indicators obtained from text. The experimental results also demonstrate the high accuracy of the proposed model compared with state-of-the-art prediction models.

Can Environmental Information Disclosure Enhance Firm Value?-An Analysis Based on Textual Characteristics of Annual Reports

This study examines the impact of environmental information disclosure quality on firm value for Chinese listed companies in heavily polluting industries from 2010 to 2021. By controlling for the level of leverage, growth, and corporate governance, a fixed effects model is constructed to test this relationship. Furthermore, this study analyzes the moderating effects of annual report text features, such as length, similarity, and readability, on the relationship between environmental information disclosure and firm value and the heterogeneous impact of firm ownership on this relationship. The main findings of this study are as follows: There is a positive correlation between the level of environmental information disclosure and firm value for Chinese listed companies in heavily polluting industries. Annual report text length and readability positively moderate the relationship between environmental information disclosure and firm value. Annual report text similarity negatively moderates the relationship between environmental information disclosure and firm value performance. Compared with state-owned enterprises, the impact of environmental information disclosure quality on the firm value of no-state-owned enterprises is more significant.

Predictable by publication: discovery of early highly cited academic papers based on their own features

Purpose

Predicting highly cited papers can enable an evaluation of the potential of papers and the early detection and determination of academic achievement value. However, most highly cited paper prediction studies consider early citation information, so predicting highly cited papers by publication is challenging. Therefore, the authors propose a method for predicting early highly cited papers based on their own features.

Design/methodology/approach

This research analyzed academic papers published in the Journal of the Association for Computing Machinery (ACM) from 2000 to 2013. Five types of features were extracted: paper features, journal features, author features, reference features and semantic features. Subsequently, the authors applied a deep neural network (DNN), support vector machine (SVM), decision tree (DT) and logistic regression (LGR), and they predicted highly cited papers 1–3 years after publication.

Findings

Experimental results showed that early highly cited academic papers are predictable when they are first published. The authors’ prediction models showed considerable performance. This study further confirmed that the features of references and authors play an important role in predicting early highly cited papers. In addition, the proportion of high-quality journal references has a more significant impact on prediction.

Originality/value

Based on the available information at the time of publication, this study proposed an effective early highly cited paper prediction model. This study facilitates the early discovery and realization of the value of scientific and technological achievements.

Green Financial Health Risk Early Monitoring of Commercial Banks Based on Neural Network Model in a Small Sample Environment

Financial innovations emerge in an endless stream, and it is difficult for the regulatory measures and efforts of banks in various countries and the credit risk management level of commercial banks themselves to adapt to the increasingly complex risk environment faced by banks. In the process of building GFR (green financial risk) mixed governance model, the division of powers and responsibilities of governance subjects should be effectively defined. Therefore, it is very necessary to comprehensively and systematically study and grasp the characteristics, performance, and causes of commercial banks' GFR and build an early-warning model of commercial banks' GFR to comprehensively monitor the risks of banks, so as to reduce risks and avoid crises. Therefore, this paper uses the forward three-layer BPNN (BP neural network) technology to establish a real-time warning model of commercial banks' GFR. IL (input layer) to HL (hidden layer) adopts Sigmoid function, while HL to OL (output layer) function adopts linear function Purelin function. The results show that the test result of this method is greatly improved compared with the traditional method, and the correct rate is increased from 81.27% to 94.38%. It shows that the model in this paper has achieved a good warning effect of GFR for commercial banks.

Creation, evaluation, and optimization of a domain-based dictionary

This work creates, evaluates, and optimizes a domain-based dictionary using labeled domain documents as the input. The dictionary is created using selected unigrams and bigrams from the labeled text documents. Dictionary is evaluated using the Naïve Bayes classification model. Classification Accuracy obtained is used as a metric to evaluate the effectiveness of the dictionary. The paper also studies the impact of applying the Stochastic Gradient Descent (SGD) technique, with Lasso and Ridge Regularization, on the effectiveness of a domain-based dictionary. Both, Lasso and Ridge regularization, with Ridge faring better than Lasso, help to optimize the dictionary size, without any significant reduction in the accuracy. The created dictionaries are evaluated on the dataset used for their creation and subsequently on an unseen dataset as well. The applicability of a created dictionary to classify the documents belonging to a different dataset gives an idea about the generality of that dictionary. The paper establishes that the dictionaries created using the above methodology are generic enough to classify documents of other unseen datasets.

Neural Network Technology-Based Optimization Framework of Financial and Management Accounting Model

Traditional financial accounting has gradually evolved into management accounting in order to adapt to changing times and developments. To avoid being obliterated by the times, accountants must gradually improve their professional and comprehensive abilities in order to create greater value for businesses in the AI (Artificial Intelligence) era. This article presents an AI-based financial management optimization design and proposes an AI-based accounts receivable management optimization framework based on the existing information system. A typical financial distress early-warning model is built using the BPNN (BP Neural Network) model, and the training samples of listed companies' financial data are processed iteratively using the neural network algorithm to realize the visual modeling of the object-oriented neural network and learn the training samples. Finally, the network's ability to provide early warning is put to the test. The results show that BPNN's prediction accuracy is significantly higher than that of other types, especially after years of data, with prediction results exceeding 90%. The results show that the BPNN-based financial early-warning method is feasible.

Sentiment lexicon construction for Chinese book reviews based on ultrashort reviews

Purpose

Sentiment lexicon is an essential resource for sentiment analysis of user reviews. By far, there is still a lack of domain sentiment lexicon with large scale and high accuracy for Chinese book reviews. This paper aims to construct a large-scale sentiment lexicon based on the ultrashort reviews of Chinese books.

Design/methodology/approach

First, large-scale ultrashort reviews of Chinese books, whose length is no more than six Chinese characters, are collected and preprocessed as candidate sentiment words. Second, non-sentiment words are filtered out through certain rules, such as part of speech rules, context rules, feature word rules and user behaviour rules. Third, the relative frequency is used to select and judge the polarity of sentiment words. Finally, the performance of the sentiment lexicon is evaluated through experiments.

Findings

This paper proposes a method of sentiment lexicon construction based on ultrashort reviews and successfully builds one for Chinese books with nearly 40,000 words based on the Douban book.

Originality/value

Compared with the idea of constructing a sentiment lexicon based on a small number of reviews, the proposed method can give full play to the advantages of data scale to build a corpus. Moreover, different from the computer segmentation method, this method helps to avoid the problems caused by immature segmentation technology and an imperfect N-gram language model.

Research on the effect of government media and users' emotional experience based on LSTM deep neural network

Different government media have different communication effects and users' emotional experience. It carries on a comparative research on government media selecting three different types of government media which include China’s Police Online, Central Committee of the Communist Youth League, and China’s Fire Control in the context of public health emergencies. Based on the deep learning technique, the emotion classification model of long-term memory network is constructed to analyze the emotion of the users’ comments of different government media; taking the number of contents, the number of retweets, the number of praises, and the number of comments as evaluating indicators to do comparative analysis to cross platform government medias. Through the comparative results, it is found that different types and platforms of government media have great differences in users’ emotional experience; the emotion performance of users’ comments is strongly related to the information communication power and effectiveness of government media.