School of Business and Management Department of Accounting 161 Deep Learning in Natural Language Processing Supervisor: HUANG, Allen / ACCT Co-supervisor: YANG, Yi / ISOM Student: LAU, Tsz Ching Rachel / COMP Course: UROP1100, Spring In this project, we aim to conduct sentiment analysis at the entity level for financial news, utilising deep learning and natural language processing (NLP) technology through large language models (LLM). This will be achieved by manual annotations of entity and sentiment information for excerpts of financial news articles, then consolidating that data with others and using it to train and fine-tune a language model, which is based on Bidirectional Encoder Representations from Transformers (BERT), a powerful LLM. Finally, the model is asked to predict entity-sentiment labels for a test dataset, which is then compared with the actual human-annotated values. Through an evaluation of the results, we conclude that our model is able to give relatively good predictions. Deep Learning in Natural Language Processing Supervisor: HUANG, Allen / ACCT Co-supervisor: YANG, Yi / ISOM Student: LEI, Sze Wing / ECON Course: UROP1100, Fall Financial fraud in public listed companies has been a huge concern in the investment field as it can incur large financial loss, for most of the cases, investors focused on numerical indicators in financial statement like profitability. Yet they often appear as a delayed detector. Our team view the problem of fraud detection from the textual perspective and examine if the MD&A report is informative to provide any pattern for predicting financial fraud. So far, our research achieves a AUC score up to 55% (the highest among all the models we trained) which might not seem to be very well-performed, yet still able to provide some insight in understanding the company’s financial behavior and suggest a possibility of future research with other indicators. Deep Learning in Natural Language Processing Supervisor: HUANG, Allen / ACCT Co-supervisor: YANG, Yi / ISOM Student: LI, Qiuru / IS Course: UROP1100, Fall In this paper, I use a deep learning algorithm – neural network to conduct a natural language processing (hereafter NLP) binary classification task for analyzing financial-fraud text. Firstly, we search for regulatory announcements and enforcement bulletin from HKEX news to define fradulent companies and to extracted their MD&A reports, then we organized the sentences from the reports with labels and reporting time. In the model building and fraud detection step, I used different kinds of neural netword models, including neural network model with Embedding, simple Recurrent Neural Network (RNN), Long-Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) for the text classification task and did the comparision of accuracy. Our results bring implication for financial fraud detection.
RkJQdWJsaXNoZXIy NDk5Njg=