Promises made by politicians, corporate leaders, and public figures have a significant impact on public perception, trust, and institutional reputation.However, the complexity and volume of such commitments, coupled with difficulties in verifying their fulfillment, necessitate innovative methods for assessing their credibility.This paper introduces the concept of Promise Verification, a systematic approach involving steps such as promise identification, evidence assessment, and the evaluation of timing for verification.We propose the first multilingual dataset, ML-Promise, which includes English, French, Chinese, Japanese, and Korean, aimed at facilitating in-depth verification of promises, particularly in the context of Environmental, Social, and Governance (ESG) reports.Given the growing emphasis on corporate environmental contributions, this dataset addresses the challenge of evaluating corporate promises, especially in light of practices like greenwashing.Our findings also explore textual and image-based baselines, with promising results from retrieval-augmented generation (RAG) approaches. 1

Climate change is an increasing environment concern, carbon markets are received attention because they can reduce emissions by carbon trading. With the rising in sustainable finance demand, there need some innovative tools to study investors' attitude. Although there has been an increasing amount of literature on market indicator and sentiment analysis, there is a limited research focus on carbon market. Therefore, this research purpose is to develop carbon market sentiment indicator. The research method contains three mainly phase. By using pretrained models like BERT, GPT, and so on, the research can obtain sentiment analysis results from news and social media data. Then, taking a specific carbon market as the object, and the selected sentiment indicators, the carbon market response for a period can be obtained at last. Due to complexity and the lack of transparency in the institutional and financial infrastructure for carbon market transactions, there has the problem of market volatility. The contribution of research is that provides a comprehensive sentiment index for carbon market. By this way, the research expects to provide valuable insights into sustainable finance and help investors to make better decision.

Financial Documents reveal important financial information about a company's financial performance which plays a vital role not only to the stakeholders but also to the public. Therefore, many researchers utilize dynamic Text mining methods in financial document to identify, analyze, predict or evaluate a company's future financial value. In order to find deeply the relationship between companies and the stakeholders, provide a simplified method for them to identify the future financial performance of the corporation. In this paper, we present a Chinese Information Extraction System (CFIES) for Financial Knowledge Graph (FinKG). The major findings of the research show an increased importance of the key audit matters in finance. The major research contribution of this paper is that we have developed CFIES which can extract the tuples from the financial reports. The adoption of the information system can assist the development of a knowledge graph that can discover deep financial knowledge in the finance domain. The managerial implication is that building CFIES can efficiently enable us to clarify the complicated relationship between the corporations, board of directors, investors, and especially the asset, assisting the stakeholders to discover a new financial knowledge representation and to make a financial decision.

Cryptocurrency anti-money laundering has become an important research topic in recent years. Legal empirical research combined with AI technology has received considerable attention. How to construct a knowledge graph of cryptocurrency anti-money laundering in a small sample of international cases and judgments on the prevention and control of cryptocurrency money laundering has become an essential issue for a better understanding of the relationship between the crime patterns and emerging financial technologies. In this study, we proposed artificial intelligence meta-learning with a few-shot learning model to construct a cryptocurrency anti-money laundering knowledge graph. The research method of this study aims at the abuse of electronic payment tools and cryptocurrency in various crimes by analyzing the causes and background, the amount of money, the type of crime, and the growth trend in recent years. The contribution of this study is that the proposed AI cryptocurrency anti-money laundering knowledge graphs in fintech can be applied to the content analysis and question-and-answer system of legal documents.

Transfer learning and Transformer-based language models play important roles in modern natural language processing research community. In this paper, we propose Transformer model's fine-tuning and data augmentation (TMFTDA) techniques for conversational texts and noisy user-generated content. We use two NTCIR-15 tasks, namely the first Dialogue Evaluation (DialEval-1) task and the second Numeral Attachment in Financial Tweets (FinNum-2) task, to evaluate the efficacy of TMFTDA. Experimental results show that TMFTDA substantially outperforms the baselines model of Bidirectional Long Short-Term Memory (Bi-LSTM) in multi-turn dialogue system evaluation at DialEval-1's Dialogue Quality (DQ) and Nugget Detection (ND) subtasks. Moreover, TMFTDA performs to a satisfactory level at FinNum-2 with a model of Cross-lingual Language Models using a Robustly Optimized BERT Pretraining Approach (XLM-RoBERTa). The research contribution of this paper is that, we help shed some light on the usefulness of TMFTDA, for conversational texts and noisy user-generated content in social media text analytics.

In asset allocation and time-series forecasting studies, few have shed light on using the different machine learning and deep learning models to verify the difference in the result of investment returns and optimal asset allocation. To fill this research gap, we develop a robo-advisor with different machine learning and deep learning forecasting methodologies and utilize the forecasting result of the portfolio optimization model to support our investors in making decisions. This research integrated several dimensions of technologies, which contain machine learning, data analytics, and portfolio optimization. We focused on developing robo-advisor framework and utilized algorithms by integrating machine learning and deep learning approaches with the portfolio optimization algorithm by using our predicted trends and results to replace the historical data and investor views. We eliminate the extreme fluctuation to maintain our trading within the acceptable risk coefficient. Accordingly, we can minimize the investment risk and reach a relatively stable return. We compared different algorithms and found that the F1 score of the model prediction significantly affects the result of the optimized portfolio. We used our deep learning model with the highest winning rate and leveraged the prediction result with the portfolio optimization algorithm to reach 12% of annual return, which outperform our benchmark index 0050.TW and the optimized portfolio with the integration of historical data.

Robo-Advisors has been growing attraction from the financial industry for offering financial services by using algorithms and acting as like human advisors to support investors making investment decisions. During the investment planning stage, portfolio optimization plays a crucial role, especially for the medium and long-term investors, in determining the allocation weight of assets to achieve the balance between investors expectation return and risk tolerance. The literature on the topic of portfolio optimization has been offering plenty of theoretical and practical guidance for implementing the theory; however, there is a paucity of studies focusing on the applications which are designed for Robo-Advisors. In this research, we proposed a modular system and focused on integrating big data analysis, deep learning method and the Black-Litterman model to generate asset allocation weight. We developed a portfolio optimization module which takes the information from a variety of sources, such as stocks prices, investor profile and the other alternative data, and used them as input to calculate optimal weights of assets in the portfolio. The module we developed could be used as a sub-system for Robe-Advisors, which offers a customized optimal portfolio based on investors preference.

With the advent of the artificial intelligence (AI) era, the combination of AI with financial technology (FinTech) has become a development trend in the financial industry. However, deep learning (DL) on the application of automated financial management has been rarely investigated. Thus, this research focuses on the applications of FinTech and DL in asset allocation and aims to optimize investment portfolio. The best investment portfolio in index-based funds based on Taiwan's index-type security investment trust funds are the main investment targets. Time series models for DL, that is, long short-term memory, predict the increase of each investment target and find the best investment portfolio in combination with the relevant asset allocation theory. In this research, we use the Markowitz mean-variance and Black-Litterman models as our asset allocation models for robo-advisor. Results show that the Black-Litterman model has a better accumulated return performance than the Morkowitz model and outperforms other strategies. The Human-Computer Interaction (HCI) dialogue service adopts artificial intelligence markup language (AIML) and a generative model. The main contribution of this paper is that we have developed an integrated knowledge-based and generative-based models for AI conversational robo-advisor.

With the rise of social networking, many consumers are willing to particulate in discussions on the social media, and to share and to express their comments for certain products. Enterprises can analyze the consumers' preferences and the strengths and weaknesses of diverse products on the market with a large number of online reviews. However, few studies have been made at applying deep learning to the sentiment analysis on word of mouth of smart bracelet in Chinese reviews. The purpose of this research is to explore the impact of online word of mouth on the consumers of smart bracelets and propose a long short-term memory (LSTM) recurrent neural network (RNN) deep learning model to the sentiment analysis on word of mouth of smart bracelets. The experiment results of sentiment analysis on word of mouth of smart bracelet show that the 89.92% accuracy based on the deep learning is significantly higher than the 70.67% on the Naïve Bayes algorithm and the 66.01% on the support vector machine. The contributions of this paper are two-fold. First, we have proposed an approach with deep learning to sentiment analysis on the word of mouth of smart bracelets. The experimental results show that the proposed deep learning approach can effectively enhance the accuracy of sentiment analysis. Second, through the text analysis, we have constructed a sentiment dictionary called iTSBSD for the word of mouth of smart bracelets in Chinese reviews on Taobao.

Product reviews are important information sources for consumers as they make their purchasing decisions. However, some unethical firms hire fake reviewers to generate biased positive reviews to promote their product and to damage the product reputations of their competitors. From the point of view of online product review platform providers, it is essential to keep the platform neutral and unbiased by detecting fake reviews and preventing fake reviewers from spreading biased reviews. In the current study, we attempt to use temporal and sentiment analyses as cues to separate fake reviews from authentic product reviews. Real case data of fake reviews in Taiwan was used for this temporal and sentiment analysis. Based on the analysis results, we find that fake reviewers usually generated and replied to fake reviewers during normal work hours. In contrast, ordinary users only generated and replied to a small proportion of normal product reviews during work hours. They generated and replied to normal product reviews the most during off-work hours and weekends. Additionally, the current study also revealed that more than half of fake reviewers replied others' responses to their own fake reviews no later than within one day. The research results revealed that temporal and sentiment analyses have the potential to serve as cues to detect fake reviews and fake reviewers.

In recent years, along with the prevalence of social networking sites, the illustrators of Wretch have accordingly transferred to new community platform. These illustrator's fan pages have become popular through viral marketing. Many companies have spotted enormous business opportunities and then worked with illustrators to boost sales by combining illustrators and commercial products. By utilizing a research model of relationship quality, online word-of-mouth, purchase intention and perceived value, the purpose of study is to explore whether fans would purchase the product endorsed by their favorite illustrators or other peripheral products. The finding show that relationship quality is positively related to online word-of-mouth and had an indirect effect on purchase intention. Online word-of-mouth and purchase intention are positively correlated. In addition, perceived value is positively related to relationship quality, online word-of-mouth, and purchase intention. The main contribution of this research is in proposing a new research model and also discovering the importance of customer perceived value to the hedonic value of products.

Investors have always been interested in stock price forecasting. Since the development of electronic media, hundreds pieces of financial news are released on different media every day. Numerous studies have attempted to examine whether the stock price forecasting through text mining technology and machine learning could lead to abnormal returns. However, few of them involved the discussion on whether using different media could affect forecasting results. Financial sentiment analysis is an important research area of financial technology (FinTech). This research focuses on investigating the influence of using different financial resources to investment and how to improve the accuracy of forecasting through deep learning. The experimental result shows various financial resources have significantly different effects to investors and their investments, while the accuracy of news categorization could be improved through deep learning.

Consumer generated product reviews are considered as more persuasive than commercial advertising, and are now an important message source to make purchase decision. Nevertheless, firms may purposely hire spammers to create fake reviews to promote their products and to demote products of their competitors. To create the opinion majority, firms may hire a group of spammers rather than just one or few individual spammers to write fake reviewers. These spammers may act as a group to support other spammers to create a social consensus or majority of opinions. In the study, we attempt to adopt a real case to analyze the social network of spammers by K-core and Clique analysis. Our research results show that the social connection among spammers is stronger than that among non-spammers. Moreover, K-cores and cliques can be used as cues to identify spammers.

With the rapid growth of the Internet and database technologies in recent years, question answering systems (QAS) have emerged as important applications. As most evaluation models focus on system-centered evaluation, user-centered evaluation has attracted little attention. Although many QAS have been implemented, little work has been done on the development of a user-centered evaluation for QAS. User-centered evaluation is used to understand a userpsilas needs and identify important dimensions and factors in the development of an information system in order to improve its acceptance. The purpose of this study is to develop a user-centered evaluation model for QAS from the userpsilas perspective. The proposed user-centered evaluation model provides a framework for the design of question answering systems from the userpsilas perspective to enhance user satisfaction and acceptance of QAS.

Question classification (QC) plays a key role in automated question answering (QA) systems. In Chinese QC, for example, a question is analyzed and then labeled with the question type it belongs to and the expected answer type. In this paper, we propose a novel method of Chinese QC that integrates syntactic tags and semantic tags into an alignment-based approach. We adopt a template alignment (TA) algorithm to process large collections of Chinese questions and compare the classification results with those of INFOMAP, a human annotated knowledge inference engine for Chinese questions. We experimented with two approaches for the proposed system: a majority algorithm and a machine learning method that uses Support Vector Machine (SVM). The TA algorithm performs well with both approaches. The experimental results show that the accuracy achieved by TA (85.5%) is comparable to that of INFOMAP (88%). In contrast, QC based on the SVM approach, which incorporates syntactic features and TA yields an accuracy rate of 91.5%.

Question Answering (QA) research has been conducted in many languages. Nearly all the top performing systems use heavy methods that require sophisticated techniques, such as parsers or logic provers. However, such techniques are usually unavailable or unaffordable for under-resourced languages or in resource-limited situations. In this article, we describe how a top-performing Chinese QA system can be designed by using lightweight methods effectively. We propose two lightweight methods, namely the Sum of Co-occurrences of Question and Answer Terms (SCO-QAT) and Alignment-based Surface Patterns (ABSPs). SCO-QAT is a co-occurrence-based answer-ranking method that does not need extra knowledge, word-ignoring heuristic rules, or tools. It calculates co-occurrence scores based on the passage retrieval results. ABSPs are syntactic patterns trained from question-answer pairs with a multiple alignment algorithm. They are used to capture the relations between terms and then use the relations to filter answers. We attribute the success of the ABSPs and SCO-QAT methods to the effective use of local syntactic information and global co-occurrence information. By using SCO-QAT and ABSPs, we improved the RU-Accuracy of our testbed QA system, ASQA, from 0.445 to 0.535 on the NTCIR-5 dataset. It also achieved the top 0.5 RU-Accuracy on the NTCIR-6 dataset. The result shows that lightweight methods are not only cheaper to implement, but also have the potential to achieve state-of-the-art performances.

The rapid growth of the Internet and instant messaging (IM) offers new opportunities as well as challenges to both educators and students. In this paper, we propose an intelligent tutoring agent (ITA) that uses the ontology, INFOMAP, and question answering techniques through the instant messaging platform for the "operating system " course. The ITA embeds the above techniques in the teaching process and plays the role of a tutoring agent to help a teacher track, record, and understand a student's status. The ITA interface can interpret natural language to facilitate communication between the student and the tutor. Student can query and learn the concept of "operating system" through MSN Messenger, which ITA adopts as the communication protocol. The proposed ITA is accessible by adding the contact ID: [email protected] to the MSN Messenger contacts list.