A data processing method, device, and medium for acquiring data on predictive event fluctuations.

Abstract

This invention relates to the field of fluctuation data prediction technology, and in particular to a data processing method, device, and medium for acquiring fluctuation data of predicted events. The method includes: obtaining a target feature vector of a first object based on first object data; obtaining a target priority of the first object based on the first object data, so as to select a second object based on the target priority of the first object; obtaining a first event influence feature vector corresponding to the second object from the target feature vector of the first object, so as to obtain the second event influence degree of the second object based on the feature vector corresponding to the second object; obtaining the target predicted value of the second object based on the second event influence degree of the second object. Through an event triggering mechanism, it can be seen that false predictions with no event and hard-given direction are eliminated, significantly improving the short window prediction accuracy; multi-source heterogeneous signal fusion and online feedback calibration realize end-to-end full-process automation, greatly reducing manual intervention.

Description

Technical Field

[0001] This invention relates to the field of fluctuation data prediction technology, and in particular to a data processing method, device and medium for acquiring fluctuation data of predicted events. Background Technology

[0002] Prediction markets, as important platforms for probabilistic pricing of events that have not yet occurred (such as election results, economic indicators, product launches, etc.), have their market prices (usually represented as probability values ​​between 0 and 1) considered as real-time aggregations of collective wisdom, possessing significant decision-making reference value. However, the events contracts in prediction markets inherently possess characteristics such as strong event-driven nature, discrete and volatile prices, and abundant noise information, posing significant technical challenges to the accurate prediction of their short-term prices.

[0003] Currently, existing technologies have the following main shortcomings: Insufficient synergistic fusion of multi-source signals: Most existing prediction methods rely on single time-series statistical models (such as LSTM) or shallow sentiment analysis of news texts. They fail to effectively integrate various heterogeneous signals that comprehensively influence market price predictions, including news text streams (event semantics), order book data (market depth and liquidity), on-chain fund flows (large position changes, applicable to blockchain prediction markets), and social media sentiment (heat and emotion). This fragmented approach prevents models from forming a comprehensive and accurate perception of market conditions.

[0004] The quantification mechanism for event impact is crude: Traditional methods lack sophisticated quantitative models when dealing with event-driven price changes. This is specifically manifested in: Lack of initial event identification: It is impossible to effectively distinguish between initial major news and repeated or secondary information, and it is easy to repeatedly weight multiple reports of the same event, leading to overreaction.

[0005] Lack of "Price-In": It is impossible to determine whether the impact of an event has been partially or fully priced in by the current market price, causing the model to output an excessively large impact some time after the event occurs.

[0006] Ignoring differences in event types: Failure to distinguish between "main events" that directly affect the final settlement results of the target (such as "the two countries formally signing an agreement") and "secondary events" that only produce short-term emotional impact (such as "the two sides conducting informal contacts") leads to a mismatch in the axis of event impact assessment.

[0007] Short-window forecasts suffer from poor stability: For forecasts with short time windows such as 30 minutes and 60 minutes, traditional time series models are highly susceptible to market random noise, generating a large number of false forecast signals. Especially during the "information vacuum period" when no major events occur, the model will still output unfounded directional forecasts based on the inertial trends of historical data ("giving direction without events"), which seriously reduces the credibility and practicality of the forecast results.

[0008] The system suffers from low levels of automation and intelligence: From target selection and event analysis to prediction result generation, existing technical solutions still require significant manual intervention, such as manually selecting valuable targets and manually interpreting the impact of news events. This is not only inefficient and unable to handle the prediction needs of hundreds or thousands of concurrent targets, but also results in a lack of closed-loop feedback and adaptive calibration capabilities throughout the prediction process.

[0009] Therefore, how to develop a short-term price forecasting method for the market that can deeply integrate multi-source heterogeneous signals, finely quantify the impact of events, and possess high stability and full-process automation has become a key technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0010] According to a first aspect of the present invention, a data processing method for acquiring predicted event fluctuation data is provided, comprising the following steps: S1, based on m first object data A={A1, ..., A...} i , ..., A m}, obtain the target feature vector F = {F1, ..., F2} of the first object. i , ..., F m}, where A i It is the data of the i-th object and A i ={A i1 A i2 A i3 A i4}, A i1 The data is provided by the first type of data source, A. i2 The data is provided by the second type of data source, A. i3 The data is provided by a third type of data source, A. i4 This is data provided by the fourth type of data source; F i It is A i The first event affecting the feature vector of the first object; the value of i ranges from 1 to m; Among them, the first to fourth data sources all provide different types of data; S2, based on A, obtain the target priority S of the first object: S = {S1, ..., S...} i S m}, so that the second object can be selected based on S; S3, obtain the first event influence feature vector corresponding to the second object from F, so as to obtain the second event influence degree of the second object based on the feature vector corresponding to the second object; S4. Obtain the target predicted value of the second object based on the impact degree of the second event on the second object.

[0011] According to a second aspect of the present invention, a non-transitory computer-readable storage medium is provided, wherein at least one instruction or at least one program is stored therein, the at least one instruction or the at least one program being loaded and executed by a processor to implement the above-described data processing method for acquiring predictive event fluctuation data.

[0012] According to a third aspect of the present invention, an electronic device is provided, including a processor and the aforementioned non-transitory computer-readable storage medium.

[0013] The present invention has at least the following beneficial effects: This invention discloses a data processing method for acquiring predictive event fluctuation data, comprising: based on m first object data A = {A1, ..., A...} i , ..., A m}, obtain the target feature vector F = {F1, ..., F2} of the first object. i , ..., F m}, where A i It is the data of the i-th object and A i ={A i1 A i2 A i3 A i4}, A i1 The data is provided by the first type of data source, A. i2 The data is provided by the second type of data source, A. i3 The data is provided by a third type of data source, A. i4 This is data provided by the fourth type of data source; F i It is A i The first event influence feature vector of the corresponding first object; the value of i ranges from 1 to m; where the data types provided by the first to fourth data sources are all different; according to A, the target priority S of the first object is obtained as S = {S1, ..., S...} i S m}, so that based on S, the second object is selected; the first event influence feature vector corresponding to the second object is obtained from F, so as to obtain the second event influence degree of the second object according to the feature vector corresponding to the second object; Based on the impact of the second event on the second object, the target prediction value of the second object is obtained. The event triggering mechanism eliminates false predictions with no event and no hard direction, significantly improving the accuracy of short window prediction. Multi-source heterogeneous signal fusion and online feedback calibration realize end-to-end full-process automation, greatly reducing manual intervention. Output confidence and display radius, coupled with bilingual causal explanation, enhance the transparency and reliability of decision-making. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart of a data processing method for obtaining predicted event fluctuation data provided in an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] This invention provides a data processing method for obtaining predicted event fluctuation data, the method comprising the following steps: S1, based on m first object data A={A1, ..., A...} i , ..., A m}, obtain the target feature vector F = {F1, ..., F2} of the first object. i , ..., F m}, where A i It is the data of the i-th object and A i ={A i1 A i2 A i3 A i4}, A i1 The data is provided by the first type of data source, A. i2 The data is provided by the second type of data source, A. i3 The data is provided by a third type of data source, A. i4 This is data provided by the fourth type of data source; F i It is A i The first event of the corresponding first object affects the feature vector; the value of i ranges from 1 to m.

[0018] Specifically, the first object is an object used for trading or pricing.

[0019] The data provided by the first to fourth data sources are all of different types; for example, the first data source provides text from news platforms; the second data source provides order data tables; the third data source provides on-chain fund flow data tables; and the fourth data source provides text from trending platforms.

[0020] Specifically, step S1 also includes the following steps to obtain F. i : S11, A i1 The input is fed into a pre-defined large speech model to obtain A. i1 The corresponding eigenvector F i1 ; S12, A i2 The input is fed into the first preset feature extraction model to obtain A. i2 The corresponding eigenvector F i2 ; S13, A i3 The input is fed into the second preset feature extraction model to obtain A. i3 The corresponding eigenvector F i3 ; S14, A i4 The input is fed into the third preset feature extraction model to obtain A. i4 The corresponding eigenvector F i4 ; S15, F i1 F i2 F i3 and F i4 F is fused through the cross-attention mechanism. i .

[0021] Furthermore, F i1 F i2 F i3 and F i4 The vector dimensions are inconsistent.

[0022] S2, based on A, obtain the target priority S of the first object: S = {S1, ..., S...} i S m}, so that the second object can be selected based on S.

[0023] Specifically, in step S2, S i The following conditions must be met: S i =W1×P i1 +W2×Pi2 +W3×P i3 +W4×P i4 +W5×P i5 Among them, P i1 It is A i The corresponding volatility of the first object, P i2 It is A i The first heat of the corresponding first object, P i3 It is A i The probability of the first object, P i4 It is A i The proximity of the corresponding first object, P i5 It is A i The corresponding weights are: W1 is the volatility weight of the first object, W2 is the primary popularity weight of the first object, W3 is the probability weight of the first object, W4 is the proximity weight of the first object, and W5 is the secondary popularity weight of the first object. For example, the volatility of the first object is its price volatility, its primary popularity is its market popularity, its probability is its suspense level, and its secondary popularity is its news popularity. Further explanation: the price volatility of the first object is the extreme value and percentile difference of its price difference within 48 hours; the market popularity of the first object is the increase in its trading volume and open interest within 24 hours; the suspense level of the first object is the inversely normalized value of its current price from 0.5; the proximity of the first object is the inversely normalized value from its pricing deadline; and the news popularity of the first object is the ratio of the number of related news items to the number of first-release events.

[0024] The further specific definitions are shown in the following formula, and the meanings of each variable are as follows: ; Where, P i1 P represents the price volatility of the i-th first object. i (t) represents the order book price of the i-th first object at time t; t0 represents the current time node; P Q3 i With P Q1 i These are the upper and lower quartiles of the order book price sequence within a 48-hour window for the object; the numerator represents the extreme value of the price difference within 48 hours, and the denominator represents the quantile difference of the price difference. The ratio of the two eliminates the influence of units.

[0025] ; Among them, P i2 V represents the market popularity of the i-th first object; 24h iThis refers to the cumulative trading volume of this object over the past 24 hours; △H 24h i V represents the absolute value of the net change in the open interest of this object over 24 hours. base i The baseline trading volume for this object (the average of the past 7 days) is used for normalization.

[0026] ; Among them, P i3 P represents the suspense level of the i-th first object; i (t0) is the order book price of the object at the current time point, and its value ranges from [0, 1]; |P i (t0)-0.5| represents the distance of the current order book price from 0.5; through The inverse normalization makes P more stable as the current price approaches 0.5. i3 The closer P is to 1 (the greater the suspense), the closer it is to 0 or 1. i3 The closer it is to 0, the less suspense there is.

[0027] Among them, P i4 T represents the proximity of the i-th first object; end i t0 is the pricing deadline for the i-th first object; t0 is the current time; T base i The baseline time length for this object (the total duration from creation to expiration) is used for normalization; P i4 The value range is [0, 1], and the closer to the deadline, the greater the value of P. i4 The closer it is to 1.

[0028] Among them, P i5 N represents the news popularity of the i-th first object; news i N represents the number of news items associated with this object. first i This is the number of first events for this object; To prevent small constants with zero denominators (default) This ratio measures the density of coverage for each first-run event, reflecting the news's popularity.

[0029] Furthermore, W1+W2+W3+W4+W5=100, for example, W1=35, W2=25, W3=20, W4=15, W5=5.

[0030] In one specific embodiment, the method further includes the step of: S21, Obtain the preset object pool capacity threshold K0 and minimum time interval T. minand the preset accuracy threshold G0; S22, obtain S i Ranking value K i A i The corresponding time T between the uploaded data and the target deadline i and A i The accuracy G of the predicted value in the previous first time window t-1 i（t-1） ; S23, when K i ≤K0 and T i ≤T min And G i（t-1） When S ≥ G0, i When it is the maximum value, S i The corresponding first object is used as the second object; otherwise, it will be used as G. i（t-1） When G is at its maximum value, i（t-1） The corresponding first object becomes the second object.

[0031] S3, obtain the first event influence feature vector corresponding to the second object from F, so as to obtain the second event influence degree of the second object based on the feature vector corresponding to the second object.

[0032] In one specific embodiment, step S3 further includes the following steps: S31, parse the first initial text of the second object to obtain the first narrative path and the second narrative path of the second object; Further explanation: The first initial text is the question stem text of the second object (containing key information such as the event description, settlement conditions, and deadline of the target); the first initial text is semantically parsed and causal chain extracted by a preset large language model to obtain the first narrative path (i.e., the core causal path that determines the final settlement result of the second object, for example, when the question stem is whether the two countries sign a permanent peace agreement, the first narrative path is the path of the official agreement reached by the two countries) and the second narrative path (i.e., the related causal path that has a short-term impact on the betting price but does not directly determine the settlement, such as congressional motions, temporary ceasefires, diplomatic visits, etc. under the same topic); the output format of the large language model is preset to structured JSON, which contains three fields: path identifier, key entity, and triggering condition. After parsing, the first narrative path and the second narrative path are obtained.

[0033] S32, Obtain the second initial text set E={E1, ..., E2} of the second object. j , ..., E n};E j It is the j-th second initial text of the second object, where j ranges from 1 to n, and n is the number of second initial texts within a preset time window; where the second initial text is the news text of the second object; S33, based on all E j Get the current event label corresponding to the second object; to further explain, for each E j Constructing similarity between E and historical event texts, j The corresponding descriptive event is labeled as tag 1; otherwise, it is labeled as tag 0.

[0034] S34, when the current event label is a first-class label, the first narrative path of the second object, the second narrative path of the second object, the second initial text set of the second object, the first event influence feature vector corresponding to the second object, and the fluctuation data of the current second object are processed through a large speech model to obtain the key event influence set corresponding to the second object; furthermore, when the current event label is 1, E j The corresponding descriptive event is taken as the new event; where the key event influence set corresponding to the second object is I={I1, ..., I...} j , ..., I n}, I j It is E j The corresponding critical time impact.

[0035] Further explanation: The large language model processes each second initial text E j When scoring, the input consists of three types of information: (a) the first and second narrative paths of the second object, used to constrain the semantic alignment of the scoring with the main / branch lines; (b) the second initial text set E of the second object and the first event impact feature vector corresponding to the second object, used to provide event context; and (c) the fluctuation data of the current second object, used to allow the large language model to assess whether the event impact has been absorbed by the market (i.e., the degree of priced-in). The large language model scores E according to three dimensions: influence of the main line direction, intensity level, and timeliness. j Output the critical event impact value I in integer form. j The value ranges from -10 to +10, where a positive value indicates an upward push in the order book price, and a negative value indicates a downward push. The absolute value reflects the intensity of the impact. It also outputs the priced-in discount factor (i.e., the event impact factor) η. j This represents the proportion of the event's impact that has been priced in by the market; the final E j The corresponding critical event impact is denoted as I. j All second initial text E j Corresponding I j The set of impact values ​​of key events is I = {I1, ..., I...} j , ..., I n}

[0036] Where, LLM() represents a pre-defined large language model; Ej For the j-th second initial text; F i The first event-affected feature vector of the second object; and These represent the first narrative path (main storyline) and the second narrative path (branch storyline) of the second object, respectively; P0 represents the current fluctuation data (market price) of the second object.

[0037] S35, based on the key event influence set corresponding to the second object, obtain the second event influence of the second object; wherein, the method of obtaining the key influence of the historical events corresponding to the second object is consistent with the key influence of the current events corresponding to the second object.

[0038] Specifically, in step S35, the second event influence degree I0 of the second object: , among which, I j The key event impact concentration E corresponding to the second object j The corresponding critical event impact, η j It is the event impact factor and η j ∈[0,1],t j It is E j The corresponding event time point, t0 is the current time point.

[0039] S4. Obtain the target predicted value of the second object based on the impact degree of the second event on the second object.

[0040] Specifically, in step S4, the target prediction value P of the second object... φ : In this context, clip() is the constraint function, α is the first influence factor, β is the second influence factor, γ is the third influence factor, P0 is the current fluctuation data, φ is the current time window length, φ0 is the preset time window length threshold, and V φ It is the volatility of the fluctuation data within the φ time window.

[0041] Furthermore, φ∈{φ 0 1, φ 0 2, φ 0 3}, φ 0 1 = 30 (min), φ 0 2 = 60 (min), φ 0 3 = 90 (min).

[0042] In one specific embodiment, the method further includes the step of: S5, Determine the target prediction mode type of the second object based on the target prediction value of the second object; S7. Based on the target prediction value of the second object, obtain the target verification data of the target prediction value.

[0043] Specifically, step S5 also includes the following steps: S51, when |P φ -P0|≥θ φ And V φ <V max It directly outputs the predicted value of the target; where θ φ V is the threshold corresponding to φ. max It is the upper limit of volatility for volatility data within the φ time window; S52, when |P φ -P0|≥θ φ And V φ ≥V max Direct trigger mode with amplitude limiting; S53, when |P φ -P0|<θ φ And I j ×(1-η j When ≠ 0, continue the mode; S54, when P φ When =P0, there is no data mode.

[0044] Specifically, steps S5 and S7 also include step S6: based on the target prediction value of the second object, output the confidence level and display radius of each prediction window.

[0045] Specifically, step S6 also includes the following steps: S61, Obtain the preset prediction window set e = {e1, e2, ..., e...} k}, output confidence level Ce, which satisfies the following condition: Ce=σ(μ1×ΔI+μ2×Ve+μ3×De), where Ce∈[0,1]; where ΔI is the cumulative event impact magnitude corresponding to the second object, Ve is the volatility of the fluctuation data under time window e, De is the priced-in degree of the second object, μ1, μ2, and μ3 are the preset first weight, second weight, and third weight, respectively, and σ() is the Sigmoid function; S62, Based on the confidence level Ce, determine the display radius Re. The display radius and confidence level are inversely proportional: Re = R max ×(1-e), where R max This is the preset maximum display radius; further explanation: the higher the confidence level Ce, the narrower the display radius Re. S63, the prediction results are presented to the user in the form of a front-end interval, wherein the front-end interval is [Pe-Re, Pe+Re], where Pe is the target prediction value of the second object under the prediction window e, and the uncertainty of the prediction is intuitively conveyed through the interval; S64, invoke the preset large language model, based on the key event impact set of the second object and the target prediction value P. δ和 Confidence level C δ Generate explanatory text, which is bilingual in Chinese and English.

[0046] In one specific embodiment, step S7 further includes the following step: S71, obtain the target periodic back-calculation accuracy U(e) and mean absolute error MAE(e); U(e) satisfies the following conditions:

[0047] Among them, MAE(e) meets the following conditions: .

[0048] S74, when U(e)≥U 0 e When U(e) is in a certain state, the target is retained in the candidate set for display; when U(e) is in a certain state, the target is retained in the candidate set for display. 0 e At that time, the target will automatically be removed from the best display position; U 0 e It is the set threshold.

[0049] The above embodiment provides a data processing method for obtaining predicted event fluctuation data, including: based on m first object data A={A1, ..., A...} i , ..., A m}, obtain the target feature vector F = {F1, ..., F2} of the first object. i , ..., F m}, where A i It is the data of the i-th object and A i ={A i1 A i2 A i3 A i4}, A i1 The data is provided by the first type of data source, A. i2 The data is provided by the second type of data source, A. i3 The data is provided by a third type of data source, A. i4 This is data provided by the fourth type of data source; F i It is A i ​The first event influence feature vector of the corresponding first object; the value of i ranges from 1 to m; where the data types provided by the first to fourth data sources are all different; according to A, the target priority S of the first object is obtained as S = {S1, ..., S...} i S m}, so that based on S, the second object is selected; the first event influence feature vector corresponding to the second object is obtained from F, so as to obtain the second event influence degree of the second object according to the feature vector corresponding to the second object; Based on the impact of the second event on the second object, the target prediction value of the second object is obtained. The event triggering mechanism eliminates false predictions with no event and no hard direction, significantly improving the accuracy of short window prediction. Multi-source heterogeneous signal fusion and online feedback calibration realize end-to-end full-process automation, greatly reducing manual intervention. Output confidence and display radius, coupled with bilingual causal explanation, enhance the transparency and reliability of decision-making.

[0050] Embodiments of the present invention also provide a non-transitory computer-readable storage medium that can be disposed in an electronic device to store at least one instruction or at least one program related to implementing a method in the method embodiments, wherein the at least one instruction or the at least one program is loaded and executed by the processor to implement the method provided in the above embodiments.

[0051] Embodiments of the present invention also provide an electronic device, including a processor and the aforementioned non-transitory computer-readable storage medium.

[0052] While specific embodiments of the invention have been described in detail by way of example, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of the invention. It should also be understood that various modifications can be made to the embodiments without departing from the scope and spirit of the invention. The scope of the invention is defined by the appended claims.

Claims

1. A data processing method for acquiring fluctuation data of predicted events, characterized in that, The method includes the following steps: S1, based on m first object data A={A1, ..., A...} i , ..., A m }, obtain the target feature vector F = {F1, ..., F2} of the first object. i , ..., F m }, where A i It is the data of the i-th object and A i ={A i1 A i2 A i3 A i4 }, A i1 The data is provided by the first type of data source, A. i2 The data is provided by the second type of data source, A. i3 The data is provided by a third type of data source, A. i4 This is data provided by the fourth type of data source; F i It is A i The first event affecting the feature vector of the first object; the value of i ranges from 1 to m; Among them, the first to fourth data sources all provide different types of data; S2, based on A, obtain the target priority S of the first object: S = {S1, ..., S...} i , ..., S m }, so that the second object can be selected based on S; S3, obtain the first event influence feature vector corresponding to the second object from F, so as to obtain the second event influence degree of the second object based on the feature vector corresponding to the second object; S4. Obtain the target predicted value of the second object based on the impact degree of the second event on the second object.

2. The data processing method for acquiring predicted event fluctuation data according to claim 1, characterized in that, Step S1 also includes the following steps to obtain F. i : S11, A i1 The input is fed into a pre-defined large speech model to obtain A. i1 The corresponding eigenvector F i1 ; S12, A i2 The input is fed into the first preset feature extraction model to obtain A. i2 The corresponding eigenvector F i2 ; S13, A i3 The input is fed into the second preset feature extraction model to obtain A. i3 The corresponding eigenvector F i3 ; S14, A i4 The input is fed into the third preset feature extraction model to obtain A. i4 The corresponding eigenvector F i4 ; S15, F i1 F i2 F i3 and F i4 F is fused through the cross-attention mechanism. i .

3. The data processing method for acquiring predicted event fluctuation data according to claim 1, characterized in that, In step S2, S i Meets the following conditions: S i =W1×P i1 +W2×P i2 +W3×P i3 +W4×P i4 +W5×P i5 ; Among them, P i1 It is A i The corresponding volatility of the first object, P i2 It is A i The first heat of the corresponding first object, P i3 It is A i The probability of the first object, P i4 It is A i The proximity of the corresponding first object, P i5 It is A i The corresponding second popularity weight of the first object; W1 is the volatility weight of the first object, W2 is the first popularity weight of the first object, W3 is the probability weight of the first object, W4 is the proximity weight of the first object, and W5 is the second popularity weight of the first object.

4. The data processing method for acquiring predicted event fluctuation data according to claim 1, characterized in that, The method further includes the following steps: S21, Obtain the preset object pool capacity threshold K 0, Minimum time interval T min and the preset accuracy threshold G0; S22, obtain S i Ranking value K i A i The corresponding time T between the uploaded data and the target deadline i and A i The accuracy G of the predicted value in the previous first time window t-1 i（t-1） ; S23, when K i ≤K0 and T i ≤T min And G i（t-1） When S ≥ G0, i When it is the maximum value, S i The corresponding first object is used as the second object; otherwise, it will be used as G. i（t-1） When G is at its maximum value, i（t-1） The corresponding first object becomes the second object.

5. The data processing method for acquiring predicted event fluctuation data according to claim 1, characterized in that, Step S3 also includes the following steps: S31, parse the first initial text of the second object to obtain the first narrative path and the second narrative path of the second object; S32, Obtain the second initial text set E={E1, ..., E2} of the second object. j , ..., E n };E j It is the j-th second initial text of the second object, where the value of j ranges from 1 to n, and n is the number of second initial texts under the preset time window; S33, based on all E j Get the current event tag corresponding to the second object; S34, when the current event label is the first type of label, the first narrative path of the second object, the second narrative path of the second object, the second initial text set of the second object, the first event influence feature vector corresponding to the second object, and the fluctuation data of the current second object are obtained through the large speech model to obtain the key event influence set corresponding to the second object; S35, based on the key event influence set corresponding to the second object, obtain the second event influence of the second object; wherein, the method of obtaining the key influence of the historical events corresponding to the second object is consistent with the key influence of the current events corresponding to the second object.

6. The data processing method for acquiring predicted event fluctuation data according to claim 5, characterized in that, In step S35, the second event influence degree I0 of the second object: , among which, I j The key event impact concentration E corresponding to the second object j The corresponding critical event impact, η j It is the event impact factor, t j It is E j The corresponding event time point, t0 is the current time point.

7. The data processing method for obtaining predicted event fluctuation data according to claim 1, characterized in that, In step S4, the target prediction value P of the second object w : In this context, clip() is the constraint function, α is the first influence factor, β is the second influence factor, γ is the third influence factor, P0 is the current fluctuation data, φ is the current time window length, φ0 is the preset time window length threshold, and V w It is the volatility of the volatility data within the W time window.

8. The data processing method for acquiring predicted event fluctuation data according to claim 1, characterized in that, The method further includes the following steps: S5, Determine the target prediction mode type of the second object based on the target prediction value of the second object; S7. Based on the target prediction value of the second object, obtain the target verification data of the target prediction value.

9. A non-transitory computer-readable storage medium, wherein the storage medium stores at least one instruction or at least one program segment, characterized in that, The at least one instruction or the at least one program segment is loaded and executed by the processor to implement the data processing method for acquiring predicted event fluctuation data as described in any one of claims 1-8.

10. An electronic device, characterized in that, Includes a processor and the non-transitory computer-readable storage medium as described in claim 9.

Images