CN-121501998-B - Text emotion index construction method based on macroscopic factor

Abstract

The invention discloses a text emotion index construction method based on macroscopic factors, which comprises the steps of obtaining a predefined macroscopic factor sequence, carrying out standardization processing to obtain a standardized macroscopic factor sequence, obtaining emotion values of various news events, calculating static weights of the various news events based on the standardized macroscopic factors and the emotion values of the various news events, adjusting the static weights according to the current macroscopic factor level to obtain dynamic weights of the various news events, and carrying out weighted summation on the emotion values of the various news events and the dynamic weights corresponding to the emotion values to generate the text emotion index. According to the invention, the weight of the news event text emotion is adjusted by utilizing the macroscopic factor, so that quantitative analysis and exponential expression of market emotion are realized.

Inventors

• Miao Chaohao
• YANG BO
• XU HAO
• ZHAO WENTIAN
• Feng Chuxuan
• XIE YI
• Du Zeren
• ZHANG QINJIE
• RUAN KAI
• LIN TINGMAO

Assignees

• 之江实验室
• 建信金融科技有限责任公司

Dates

Publication Date

20260512

Application Date

20260112

Claims (6)

1. 1. A method for constructing a text emotion index based on macroscopic factors, comprising: obtaining a predefined macroscopic factor sequence and carrying out standardization treatment to obtain a standardized macroscopic factor sequence; Obtaining emotion values of various news events; Based on standardized macro factors and emotion values of various news events, calculating static weights of the various news events, wherein the static weights comprise Beta coefficients of the various news events relative to the macro factor sequences, so as to represent long-term association strength of emotion of the news event category and macro economic change; wherein determining static weights for each type of news event based on Beta coefficients for each type includes: Initial risk weights are distributed based on Beta coefficient absolute values of all categories, so that expected risk contributions of all macroscopic factors to the text emotion indexes tend to be balanced, normalization processing is carried out, and static weights of all types of news events are obtained; The method also comprises the step of checking the rationality of the static weight based on the historical sample, specifically, checking whether the reaction of the emotion index constructed according to the static weight to the macroscopic factor change is balanced or not by observing; the static weight is adjusted according to the current macroscopic factor level to obtain the dynamic weight of various news events, which comprises the following steps: When the macroscopic factor level changes, the weight adjustment coefficient of the corresponding state is called to correct the static weight of each event category so as to obtain dynamic weight; And carrying out weighted summation on emotion values of various news events and dynamic weights corresponding to the emotion values to generate a text emotion index.
2. 2. The method of claim 1, wherein the macro factors include a growth factor that characterizes economic growth, a swell factor that characterizes swell of a currency, a interest rate factor that characterizes market interest rate level, a credit factor that characterizes credit environment, a rate factor that characterizes currency rate, and a liquidity factor that characterizes liquidity; the macro factor sequence is determined by constructing and principal component analyzing financial market data.
3. 3. The method for constructing a text emotion index based on macroscopic factors as recited in claim 1, wherein the news event types include employment and economic growth events, inflation related events, money policy events, financial risk events and geopolitical events, and news texts are mapped to the categories through preset event ontology, type tags of the news events are identified by using a natural language processing algorithm, and emotion values of the news events are calculated.
4. 4. The method for constructing a text emotion index based on macroscopic factors as recited in claim 1, wherein the acquisition of dynamic weights further comprises: and (3) inputting the current macroscopic factor value by adopting a neural network attention mechanism, and directly calculating the dynamic weight of each event category so as to enable the text emotion index to generate self-adaptive weight adjustment on macroscopic environment change.
5. 5. An electronic device comprising a memory and a processor, wherein the memory is coupled to the processor, wherein the memory is configured to store program data, and wherein the processor is configured to execute the program data to implement a macroscopic factor based text emotion index construction method as recited in any of claims 1-4.
6. 6. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements a macroscopic factor based text emotion index construction method as claimed in any of claims 1-4.

Description

Text emotion index construction method based on macroscopic factor Technical Field The invention relates to the field of macro economic analysis and natural language processing, in particular to a text emotion index construction method based on macro factors. Background In recent years, the relationship between media news moods and the financial market has received widespread attention. Early studies showed that the mood of media reports could have an impact on market price. For example, tetlock (2007) analyzed the text emotion of the "daily wall newspaper" monorail article, and found that pessimistic emotion therein was able to predict short-term downgoing pressures of the stock market, but this tendency to drop quickly returns to basal levels. Manela and Moreira (2017) construct a news-text-based uncertainty index by using the news of the first edition of the daily news of the street in the middle of the street in 1890, wherein the uncertainty index is remarkably increased when events such as stock market breakout, policy uncertainty period, world war, financial crisis and the like occur. The classical method generally adopts rules based on emotion dictionary counting or pre-fixed to calculate emotion indexes, and often equally weights different news texts or adopts static weight setting, so that the regulating effect of macroscopic economic environment change on news emotion influence is not considered. This results in a lack of sensitivity of the emotion index to macroscopic economic state changes, and it is difficult to reflect different impact levels of news events on market emotion in time for different economic cycles. There are also some automated means in the prior art for emotion analysis of news text. For example, chinese patent CN103793371a proposes a news text emotion tendency analysis method, which uses an emotion dictionary to segment news text clauses, and calculates the emotion tendency entropy value of each sentence to determine the overall emotion. Another example CN104462065B discloses an analysis method of event emotion type, which determines emotion attributes of an event by identifying emotion words related to the event and weighting emotion according to the association degree with the event. These techniques focus on text emotion extraction or event emotion classification itself, while still lacking consideration of dynamic weights of event impact under different macroscopic scenarios when constructing emotion indexes. In summary, how to integrate macro economic factors into the construction of text emotion indexes flexibly adjusts the influence weights of news event emotions under different macro states, so as to improve the sensitivity of the emotion indexes to macro environment changes, and the method is a problem to be solved. Disclosure of Invention The invention aims to provide a text emotion index construction method based on macroscopic factors aiming at the defects of the prior art. The invention aims at realizing the following technical scheme that the text emotion index construction method based on macroscopic factors comprises the following steps: obtaining a predefined macroscopic factor sequence and carrying out standardization treatment to obtain a standardized macroscopic factor sequence; Obtaining emotion values of various news events; Based on the standardized macroscopic factors and emotion values of various news events, calculating to obtain static weights of the various news events; the static weight is adjusted according to the current macroscopic factor level, so that dynamic weights of various news events are obtained; And carrying out weighted summation on emotion values of various news events and dynamic weights corresponding to the emotion values to generate a text emotion index. Further, the macroscopic factors include a growth factor that characterizes economic growth, a swell factor that characterizes swell of a currency, a interest rate factor that characterizes market interest rate level, a credit factor that characterizes a credit environment, a rate factor that characterizes currency rate, and a liquidity factor that characterizes liquidity; the macro factor sequence is determined by constructing and principal component analyzing financial market data. Further, the news event types comprise employment and economic growth events, expansion related events, currency policy events, financial risk events and geopolitical events, the news event is mapped to the categories through a preset event ontology, a natural language processing algorithm is adopted to identify a news event type label, and the emotion value of the news event is calculated. Further, beta coefficients of various news events relative to the macro factor sequence are calculated to represent long-term association strength of emotion of the news event category and macro economic change, and static weights of the various news events are determined based on the Beta coefficients of the vario