Approach for Multi-Label Text Data Class Verification and Adjustment Based on Self-Organizing Map and Latent Semantic Analysis

In this paper, a new approach has been proposed for multi-label text data class verification and adjustment. The approach helps to make semi-automated revisions of class assignments to improve the quality of the data. The data quality significantly influences the accuracy of the created models, for example, in classification tasks. It can also be useful for other data analysis tasks. The proposed approach is based on the combination of the usage of the text similarity measure and two methods: latent semantic analysis and self-organizing map. First, the text data must be pre-processed by selecting various filters to clean the data from unnecessary and irrelevant information. Latent semantic analysis has been selected to reduce the vectors dimensionality of the obtained vectors that correspond to each text from the analysed data. The cosine similarity distance has been used to determine which of the multi-label text data class should be changed or adjusted. The self-organizing map has been selected as the key method to detect similarity between text data and make decisions for a new class assignment. The experimental investigation has been performed using the newly collected multi-label text data. Financial news data in the Lithuanian language have been collected from four public websites and classified by experts into ten classes manually. Various parameters of the methods have been analysed, and the influence on the final results has been estimated. The final results are validated by experts. The research proved that the proposed approach could be helpful to verify and adjust multi-label text data classes. 82% of the correct assignments are obtained when the data dimensionality is reduced to 40 using the latent semantic analysis, and the self-organizing map size is reduced from 40 to 5 by step 5.

Event Study: Advanced Machine Learning and Statistical Technique for Analyzing Sustainability in Banking Stocks

Machine learning has grown in popularity in recent years as a method for evaluating financial text data, with promising results in stock price projection from financial news. Various research has looked at the relationship between news events and stock prices, but there is little evidence on how different sentiments (negative, neutral, and positive) of such events impact the performance of stocks or indices in comparison to benchmark indices. The goal of this paper is to analyze how a specific banking news event (such as a fraud or a bank merger) and other co-related news events (such as government policies or national elections), as well as the framing of both the news event and news-event sentiment, impair the formation of the respective bank’s stock and the banking index, i.e., Bank Nifty, in Indian stock markets over time. The task is achieved through three phases. In the first phase, we extract the banking and other co-related news events from the pool of financial news. The news events are further categorized into negative, positive, and neutral sentiments in the second phase. This study covers the third phase of our research work, where we analyze the impact of news events concerning sentiments or linguistics in the price movement of the respective bank’s stock, identified or recognized from these news events, against benchmark index Bank Nifty and the banking index against benchmark index Nifty50 for the short to long term. For the short term, we analyzed the movement of banking stock or index to benchmark index in terms of CARs (cumulative abnormal returns) surrounding the publication day (termed as D) of the news event in the event windows of (−1,D), (D,1), (−1,1), (D,5), (−5,−1), and (−5,5). For the long term, we analyzed the movement of banking stock or index to benchmark index in the event windows of (D,30), (−30,−1), (−30,30), (D,60), (−60,−1), and (−60,60). We explore the deep learning model, bidirectional encoder representations from transformers, and statistical method CAPM for this research.

Summarization and Sentiment Analysis for Financial News

Data is growing at an unimaginable speed around us, but what part of it is really useful information? Business leaders, financial analysts, stock market enthusiasts, researchers etc. often need to go through a plethora of news articles and data every day, and this time spent may not even result in any fruitful insights. Considering such a huge volume of data, there is difficulty in gaining precise, relevant information and interpreting the overall sentiment portrayed by the article. The proposed method helps in conceptualizing a tool that takes financial news from selected and trusted online sources as an input and gives a summary of the same along with a basic positive, negative or neutral sentiment. Here it is assumed that the tool user is familiar with the company’s profile. Based on the input (company name/symbol) given by the user, the corresponding news articles will be fetched using web scraping. All these articles will then be summarized to gain succinct and to the point information. An overall sentiment about the company will be portrayed based on the different important features in the article about the company. Keywords: Financial News; Summarization; Sentiment Analysis.

SENTiVENT: enabling supervised information extraction of company-specific events in economic and financial news

We present SENTiVENT, a corpus of fine-grained company-specific events in English economic news articles. The domain of event processing is highly productive and various general domain, fine-grained event extraction corpora are freely available but economically-focused resources are lacking. This work fills a large need for a manually annotated dataset for economic and financial text mining applications. A representative corpus of business news is crawled and an annotation scheme developed with an iteratively refined economic event typology. The annotations are compatible with benchmark datasets (ACE/ERE) so state-of-the-art event extraction systems can be readily applied. This results in a gold-standard dataset annotated with event triggers, participant arguments, event co-reference, and event attributes such as type, subtype, negation, and modality. An adjudicated reference test set is created for use in annotator and system evaluation. Agreement scores are substantial and annotator performance adequate, indicating that the annotation scheme produces consistent event annotations of high quality. In an event detection pilot study, satisfactory results were obtained with a macro-averaged $$F_1$$ F 1 -score of $$59\%$$ 59 % validating the dataset for machine learning purposes. This dataset thus provides a rich resource on events as training data for supervised machine learning for economic and financial applications. The dataset and related source code is made available at https://osf.io/8jec2/.

Predicting stock price movement using effective Thai financial probabilistic lexicon

Predicting stock price fluctuation during critical events remains a big challenge for many researchers because the stock market is extremely vulnerable and sensitive during such time. Most existing works rely on various numerical data of related factors which can impact the stock price direction. However, very few research papers analyzed the effect of information appearing in financial news articles. In this paper, a novel probabilistic lexicon based stock market prediction (PLSP) algorithm is proposed to predict the direction of stock price movement. Our approach used the proposed thai financial probabilistic lexicon (ThaiFinLex) derived from Thai financial news and stock market historical prices. The PLSP development consists of three steps. Firstly, we constructed ThaiFinLex by extracting event terms from news articles and calculating their associated probability of increasing/decreasing values of stock prices. Then, event terms with bad prediction performance were filtered out. Finally, the stock price directions were predicted using the PLSP and the remaining effective event terms. Our results indicated that the proposed model can be used for predicting stock price movement. The performance is as high as 83.33% when PLSP is used to predict stocks from the financial sector.

How Novelty and Narratives Drive the Stock Market

'Animal spirits' is a term that describes the instincts and emotions driving human behaviour in economic settings. In recent years, this concept has been discussed in relation to the emerging field of narrative economics. When unscheduled events hit the stock market, from corporate scandals and technological breakthroughs to recessions and pandemics, relationships driving returns change in unforeseeable ways. To deal with uncertainty, investors engage in narratives which simplify the complexity of real-time, non-routine change. This book assesses the novelty-narrative hypothesis for the U.S. stock market by conducting a comprehensive investigation of unscheduled events using big data textual analysis of financial news. This important contribution to the field of narrative economics finds that major macro events and associated narratives spill over into the churning stream of corporate novelty and sub-narratives, spawning different forms of unforeseeable stock market instability.

Stock Prediction Based on News Text Analysis

With the vigorous development of information technology, the textual data of financial news have grown massively, and this ever-rich online news information can influence investors’ decision-making behavior, which affects the stock market. Thus, online news is an important factor affecting market volatility. Quantifying the sentiment of news media and applying it to stock-market prediction has become a popular research topic. In this study, a financial news sentiment lexicon and an auxiliary lexicon applicable to the financial field are constructed, and a sentiment index (SI) is constructed by defining the weight of semantic rules. Then, a comprehensive sentiment index (CSI) is constructed via principal component analysis of the sentiment index and structured stock-market trading data. Finally, these two sentiment indices are added to the generalized autoregressive conditional heteroscedastic (GARCH) and the Long short-term memory (LSTM) models to predict stock returns. The results indicate that the prediction results of LSTM models are better than those of GARCH models. Compared with general-purpose lexicons, the financial lexicons constructed in this study are more stable, and the inclusion of a comprehensive investor sentiment index improves the accuracy of measuring sentiment information. Thus, the proposed lexicons allow more comprehensive measurement of the effects of external sentiment factors on stock-market returns and can improve the prediction effect of stock-return models.