School of Business and Management Department of Accounting 160 Department of Accounting Deep Learning in Natural Language Processing Supervisor: HUANG, Allen / ACCT Co-supervisor: YANG, Yi / ISOM Student: CHAN, Ka Kiu / RMBI Course: UROP1100, Spring Entity-level sentiment analysis predicts the sentiment towards a specific entity mentioned in a given text. It is extremely useful in business context, especially in the financial domain, where understanding the sentiment towards a specific company or financial instrument is essential for making informed decisions. This paper presents a domain- and task-specific dataset using BILUO tagging and a new model for entitylevel sentiment analysis in the financial context. This model aims to identify the companies, organizations and products mentioned in a given text, and extract the corresponding sentiments towards each entity simultaneously. The proposed method leverages the power of the BERT model, where a pretrained BERT model is fine-tuned and hyperparameters are optimized. Deep Learning in Natural Language Processing Supervisor: HUANG, Allen / ACCT Co-supervisor: YANG, Yi / ISOM Student: CHENG, Lai Him / IS Course: UROP1100, Spring The report is aspired to support or provide insights into Prof. HUANG, Allen and Prof. YANG Yi’s research. The project is related to deep learning in natural language processing (NLP), to be specific, training large language model (LLM). For example, Google has introduced Bidirectional Encoder Representations from Transformer (BERT) model, which is a deep-learning-based NLP model to learn textual data. In our project, I am assigned to data annotation task and to build a deep-learning model about name-entity recognition task with sentimental analysis in financial text data. The progress of my work and my reflection will be discussed in this report. Deep Learning in Natural Language Processing Supervisor: HUANG, Allen / ACCT Co-supervisor: YANG, Yi / ISOM Student: KAMATH, Aditya / QFIN Course: UROP1100, Spring This research project draws inspiration from the widely known named entity recognition (NER) task which uses a deep learning model to tag proper noun(s) in a given piece of text and its association (person, place, etc). The purpose of this research project is to develop a transformer-based deep learning model that will be able to do two main things when given a written piece of text. Firstly, it should be able to recognize the entities present, which in the financial domain, would mainly be: companies, their products as well as major financial organizations and instruments. After which, it should be able to conduct sentiment analysis of these entities in the context of the given text. The sentiment produced by the model will be either positive, negative, or neutral for each recognised entity. This will have countless applications in the world of finance as being able to stream in vast amounts of data from sources (e.g. live-streaming of news from Twitter) and receiving timely sentiment of companies based on these news will allow users to have an edge while making decisions.
RkJQdWJsaXNoZXIy NDk5Njg=