A Machine with a Man: Augmenting AI with Human Knowledge for Sentiment Extraction from Firm Disclosures

Management Science and Information Systems Seminar（2024-01）

Title: A Machine with a Man: Augmenting AI with Human Knowledge for Sentiment Extraction from Firm Disclosures

Speaker:  Jiexin Zheng, Hong Kong University of Science and Technology

Time: Monday, February 26, 10:00-11:30 a.m., Beijing Time

Location: Room 216, Guanghua Building 2

Abstract

We develop an augmentation approach that incorporates human knowledge into artificial intelligence (AI) to improve sentiment extraction from firm disclosures. This approach uses the Loughran and McDonald (LM) dictionary to represent human knowledge and employs a word embedding AI model to adjust the sentiment strength of words over time. We evaluate our approach using firms’ 10-K reports and find that the augmented sentiment measures consistently explain abnormal market returns immediately after the release of these reports. This finding indicates that the inconsistencies observed in the literature are largely due to inaccurate sentiment measures. Furthermore, we find that our augmented sentiment measures, when used as a word reweighting scheme for machine learning, exhibit improved economic significance and explanatory power for abnormal market returns. In addition, we observe that our augmented sentiment measures help predict firms’ future earnings. Our study demonstrates that the proposed augmentation approach mitigates the impact of managers’ strategic manipulation on disclosure sentiment. By performing additional analyses, we demonstrate the robustness and applicability of our approach across various word lists. Moreover, we reveal that our augmented sentiment measures outperform conventional NLP and BERT-related models in both explaining and predicting market reactions and future earnings performance. Our work highlights the complementarity between human knowledge and AI when processing textual information in financial markets.

Bio

I am currently a PhD candidate in the Department of Information Systems, Business Statistics and Operations Management (Information Systems Stream) at the Hong Kong University of Science and Technology, advised by Prof. Rong ZHENG.

I work on multidisciplinary research that employs various techniques, such as machine learning methods, experimentation, survey, and econometric modeling, to important business problems and offer multi-faceted insights. My primary research focus is Human-AI interaction. I researched and developed methodologies that utilize the complementary strengths of human intelligence and artificial intelligence to enhance the processing of large-scale data. I also study strategies for maximizing AI's benefits (e.g. augmenting workforce efficiency) while minimizing its risks (e.g. algorithm manipulation).