School of Business and Management Department of Information Systems, Business Statistics and Operations Management 214 Deep Learning NLP in finance Supervisor: YANG Yi / ISOM Student: SUN Ruijie / FINA TAM Shing Hang Boris / COGBM Course: UROP1100, Spring UROP1100, Spring The project is about analyzing hedge words from companies’ conference calls. A hedge word in Finance means a speaker trying to reduce the risks of providing false information in their speech by including uncertainties in their sentences. For instance, an example about hedge words can be “may”, “probably”, etc. Having different hedging levels in conference calls could mean the uncertainty level of that company, hence impacting its values and prospects (negatively or positively). In our project, the aim was not correlating between hedging levels with the corporate performance; but instead to utilize the Natural Language Processing techniques to analyze the hedging levels of a particular conference call based on its speakers’ transcript. Deep Learning NLP in finance Supervisor: YANG Yi / ISOM Student: JIA Tongtong / QFIN Course: UROP1100, Summer BERT, which stands for Bidirectional Encoder Representations from Transformers, has made significant progress in helping computers completing many kinds of NLP tasks by using surrounding text from both directions to establish context. Based on the original version of BERT, there have been many domain-specific BERT models fine-tuned with a specific corpus, which are better suited and perform better when dealing with tasks in specific domains, such as FinBERT (McCormick, 2021). However, we lack some fine-tuned FinBERT models which focus on Chinese finance corpora. To address this issue, we present Chinese FinBERT which is a financial domain-specific language model fine-tuned on large-scale Chinese financial corpora. The results show that our model outperforms many current models with high effectiveness and robustness.
RkJQdWJsaXNoZXIy NDk5Njg=