Because of its high relevance to data and texts, the financial sector is one of the first to be used by artificial intelligence, and knowledge mapping techniques, which are important research directions and components of artificial intelligence technology, are rapidly entering the financial arena and are increasingly becoming the cornerstone of smart finance。
Smart question-and-answer robots based on financial knowledge mapping, precise retrieval in the future will potentially be applied as a core competency in a variety of smart financial applications, such as smart investment, smart investment, smart style control, smart passenger service, smart regulation, smart operations, etc。
Smart assistant service framework
As shown in the figure above, the smart auxiliary service framework is a “material” that captures a large amount of data, information, knowledge forms from news, financial reports, the development of industry websites, etc., and is built into a knowledge map through semantic analysis and provides advanced semantic search engines, smart questions and answers, interactive knowledge management systems, document (knowledge) collaborative systems to manage, search and use financial knowledge more effectively。
Semantic searches, smart questions and answers are key nlp techniques designed to enable users to ask questions in their natural language and to conduct in-depth semantic analyses to better understand user intentions and quickly and accurately access information in the knowledge base. On the user interface, it can be in the form of a search engine (semantic search) or a robot for questions and answers (smart questions and answers)。
Semantic retrieval project architecture
As shown in the figure, the semantic search here uses the elk architecture, i. E. By entering the knowledge map as a log source, then extracting knowledge into a front-end readable text through a series of conduits, obtaining a extraction result through the log analysis index elasticsearch, and ultimately analysis through the visual analysis interface provided by kibana。
Semantic search examples
Information obtained from searches can also be processed and categorized, as shown in figure 4, by synopsis of negative messages, hot spots, mergers and acquisitions, company announcements, studies, upstream and downstream companies, industry size, etc。
A question and answer robot
The question-and-answer robot (intellectual question-and-answer system) typically includes a question-and-answer understanding, information retrieval, and the generation of answers. Question-and-answer robots are closely related to financial knowledge mapping, which provides the expression, storage and reasoning of knowledge at the semantic level, while question-and-answer robots provide access to knowledge retrieval at the semantic level. A question-and-answer robot based on a knowledge map is more responsive to the real needs of financial operations than a text-based question-and-answer machine。
There are many cases of question-and-answer robots, such as siri and microsoft mini-ice, but the general effect is not very good. This is due to the fact that they are knowledgeable intelligence assistants with too many semantic entities that need to be identified to perform multi-wheel conversations, semantics and semantics, and the accuracy rate can be greatly increased if they are just industry assistants。
Question and answer robot running process
Smart question and answer robot effects demonstration
Question and answer robot internal principles
The technical features of the question-and-answer robot are the coding (encode) of questions and given content, and the decoded answers, as shown in figure 7, i. E. The coding of search content (query) and search object (context), the decorating of key words (query) and the location of key words。
The word-bag vector approach often adopted by question-and-answer robots has limitations, as there is usually no word vector for certain important words, such as field-specific terms or common spelling errors. This is particularly true when users use non-english or non-official languages. So smart questions and answers do not use pre-trained vectors for the new pipe, but rather use learning intent and word embedding, i. E. Using bi-lstm+crf for serialization, which is used to sort out similarities between input sentences and all intents. This means that the pre-trained word vector used in an open box will not be troubled, but will be dedicated to learning its own word for a specific domain。
The value of smart questions and answers and semantic retrieval is increasingly being valued in the financial sphere. Its main applications include smart investment, smart investment and smart passenger service. In the area of intelligent research, the day-to-day work of the researcher requires extensive information to be searched through multiple channels. With financial questions and answers and semantic access, access to information will be “just ask a question”. Moreover, semantic retrieval of returns results not only in graphic web-based information, but also in stereo-based information that can organize information from various sources and provide some analytical predictions. In the area of smart passenger service and smart investment, the application of smart question-and-answer systems is mainly a robotic passenger service, whose current role is simply to assist artificial passenger service to answer a number of frequently asked questions, but has resulted in significant savings in human costs in the passenger service sector。
Through knowledge mapping, we can better serve the business landscape and trade with more extensive knowledge。
