FinKG

project introduction

FinKG denotes Financial Knowledge Graph, which is an innovative project at the intersection of finance and natural language processing (NLP). It serves as a powerful tool for stock enthusiasts and investors by harnessing the capabilities of a knowledge graph. This system is designed to facilitate stock-related inquiries and provide insightful investment recommendations.

At its core, FinBERT leverages NLP techniques for efficient entity recognition. This means that it can understand and extract crucial information from unstructured text data, such as news articles, financial reports, and user queries. By identifying entities like company names, stock symbols, and financial metrics, it creates a structured knowledge base that forms the foundation of its recommendations.

Instructions for Use

1. Login and Logout Module

User registration function: Click send to send a verification code, which will be sent to the backend. After registration, you can log in. Unregistered users have restricted access.

2. Entity Query

After logging in, you will automatically enter the entity query page. The first thing displayed is the tutorial for using the entity query function. You can follow the tutorial to select the category of the entity you want to query. Then, enter the entity name in the input box on the far left and click query to get the result. For example, querying the type “stock_name” for “Lingyun Co., Ltd.” will give the following results. The left part shows information related to the stock type, and the right part shows the shareholders of the stock. In the query interface, you can also get the following two charts. The bar chart on the left shows the yield rates for the last three days, the last month, and the last six months, respectively. The pie chart on the right shows the shareholding and proportion of the stock’s shareholders.

3. Relationship Query

Click on the relationship query on the left to enter the following interface, which has a corresponding tutorial. We won’t go into details here but will show several query situations below. First situation: Enter only one entity and click query to get other entities related to this entity. Below is an example querying “Kweichow Moutai.” Second situation: Enter two entities and a relationship to display the ‘entity-relationship-entity’ diagram. Below is an example with “Kweichow Moutai belong Big Data.” Third situation: Enter one entity and one relationship to find other entities related to this entity by the given relationship. Below, querying “own Kweichow Moutai” shows that “Kweichow Moutai Co., Ltd.” owns the stock “Kweichow Moutai.” Click to enter the stock ranking page. The left side explains the stock ranking, including the rules and a disclaimer that the ranking is for reference only. The right side shows the top 50 stocks with the highest overall ratings in all industries, with the stock ratings displayed next to them.

4. Stock Clustering

Upon entering the clustering module, you will see a related tutorial. Then, choose the concept you want to query to obtain a relationship diagram based on the clustering rules below. The relationship diagram is divided into three clusters. Different stocks can be categorized as “Golden Cross, Death Cross, No Cross” types. Nodes are represented by “red, blue, yellow” colors, respectively, and the three clusters are uniformly represented by green nodes. The query results are as follows: This is all the related stock information for the concept “Tesla,” divided into three clusters.

5. Entity Recognition

Entity recognition allows users to input a piece of text, pass it to the trained FinBert model for prediction, and display the recognized entities in the text area on the right. Users can gain a deeper understanding of related concepts, companies, and other information. For example, inputting the sentence “Zhang Hua was appointed as the CEO of Emerging Financial Group” will produce the following recognition result.

6. Intelligent Q&A

In the intelligent Q&A section, users can enter questions in the input box and click query to get related results. The “You can ask me” section below provides a general format for query sentences. For example, querying “Who are the shareholders of 000002.SZ” will produce the following result. The right side will display the input content, and our query question will be matched to different query modes in the backend. The entity type below shows what is being queried. The query result here displays the shareholders of “000002.SZ.”

7. Additional Features

1. Personal Homepage: Mainly used to check recent activities on the webpage. The corresponding edit function can also be used normally. Just enter the modification item to make the corresponding changes in the backend database (using the unchanged user_id for modification).
2. History Records The history records section will keep a more detailed record of recent queries on the webpage, including the query category, parameters, and time. If we need to find out what was done recently, we can check here.
3. User Favorites After performing a query, you can find a favorites button in the upper right corner. Clicking it will record the favorite information in the user favorites on the left sidebar. For example, querying “Lingyun Co., Ltd.”
4. Customize Click the rotating gear on the side to customize the background image.

personal duties

I’m the Team Leader of the CSC International Enterprise Internship Project

My primary responsibility involves fine-tuning and deploying the BERT model on datasets for downstream entity recognition tasks in the financial domain.