Aspect-based Sentiment Analysis in Document -- FOMC Meeting Minutes on Economic Projection

Yifei Wang

DOI: https://doi.org/10.48550/arXiv.2108.04080

2023-04-24

Abstract:The Federal Open Market Committee within the Federal Reserve System is responsible for managing inflation, maximizing employment, and stabilizing interest rates. Meeting minutes play an important role for market movements because they provide the birds eye view of how this economic complexity is constantly re-weighed. Therefore, There has been growing interest in analyzing and extracting sentiments on various aspects from large financial texts for economic projection. However, Aspect-based Sentiment Analysis is not widely used on financial data due to the lack of large labeled dataset. In this paper, I propose a model to train ABSA on financial documents under weak supervision and analyze its predictive power on various macroeconomic indicators.

Computation and Language

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to dynamically extract aspects (such as inflation, economic growth, employment, etc.) from the Federal Open Market Committee (FOMC) meeting minutes and conduct aspect - based sentiment analysis (ABSA) in the absence of a large amount of labeled data, and then evaluate the impact of these sentiments on the predictive ability of macro - economic indicators. Specifically, the paper explores the following two core issues:

1. **Can aspects be dynamically extracted in a weakly - supervised manner in the absence of a large amount of labeled data?**
- Due to the particularity of financial data, especially the subtle differences in language and tone in the FOMC meeting minutes, traditional methods based on labeled data are difficult to be directly applied. Therefore, researchers attempt to develop a model that can effectively identify and extract various economic aspects discussed in the document with the support of a small amount of labeled data.

2. **Can the aspect - based sentiment extracted from the FOMC meeting minutes be used for economic prediction?**
- By analyzing the sentiment regarding different economic aspects in the FOMC meeting minutes, researchers hope to explore the relationship between these sentiments and actual macro - economic indicators (such as inflation rate, economic growth rate, unemployment rate, etc.), thereby verifying the value of sentiment analysis in economic prediction.

To answer these questions, researchers extended the natural language processing model at the sentence level to the document level and verified the effectiveness and predictive ability of the model through experiments. Through this method, researchers not only solved the problem of insufficient data labeling, but also demonstrated the potential application value of aspect - based sentiment analysis in economic prediction.