A futures field research report processing method and system

Abstract

The application relates to the technical field of financial data processing, in particular to a futures field investment research report processing method and system. The method comprises the following steps: obtaining multiple original investment research reports of different data sources, extracting futures varieties, publishing times, core viewpoints, multiple empty direction labels and support data in each report; dividing the multiple reports into time sequence report sets of the same variety based on the futures varieties and the publishing times, judging whether the multiple empty direction labels of adjacent reports are reversed; if not, carrying out viewpoint fusion processing on the core viewpoints and the support data of the adjacent reports to generate a consistent viewpoint abstract; if yes, extracting key assumption conditions in the conflict reports and generating a viewpoint reversal event record. The application helps to automatically extract key information in the investment research reports, identify viewpoint reversal events, improve the efficiency of investment research personnel in processing massive reports and reduce the risk of information omission.

Description

Technical Field

[0001] This application relates to the field of financial data processing technology, and in particular to a method and system for processing investment research reports in the futures field. Background Technology

[0002] In the futures market, major research institutions release a large number of research reports daily. These reports include supply and demand analysis, price forecasts, bullish and bearish opinions, and supporting data for different futures contracts, serving as important references for investors' trading decisions. With the increasing number of market participants and the diversification of information dissemination channels, dozens of independent research reports from different institutions may be received daily for the same futures contract. How to quickly extract effective information and identify changing trends in opinions from this massive volume of reports has become a crucial challenge for investment research personnel.

[0003] However, existing technologies have the following drawbacks when processing futures research reports: First, research personnel need to manually read multiple reports and extract the futures contracts, publication dates, core viewpoints, bullish / bearish directional tags, and supporting data from each report. This process is time-consuming, labor-intensive, and prone to information omissions. Second, due to the lack of means to automatically compare and determine the bullish / bearish directional tags of reports on the same contract from different times, the identification of viewpoint reversal events relies entirely on manual reading of historical reports, which is extremely inefficient. Therefore, how to solve the problems of the laboriousness and potential for information omissions in manually extracting report information, and the inefficiency of manually identifying viewpoint reversal events, are urgent technical problems to be solved in this field. Summary of the Invention

[0004] To help solve the above-mentioned technical problems, this application provides a method and system for processing investment research reports in the futures field.

[0005] Firstly, this application provides a method for processing investment research reports in the futures field, which adopts the following technical solution: A method for processing investment research reports in the futures market includes: Obtain multiple original investment research reports from different data sources; Extract the futures contracts, release date, core viewpoints, bullish / bearish directional tags, and supporting data from each original investment research report; Based on the futures product and the publication time, the multiple original investment research reports are divided into a time series report set for the same product; Determine whether the long / short direction labels of two adjacent reports in the time series report set have been reversed; If no reversal occurs, the core viewpoints and supporting data of the two adjacent reports will be merged to generate a consistent viewpoint summary; If a reversal occurs, extract the two conflict reports before and after the reversal time, and analyze the key assumptions in the supporting data of the two conflict reports respectively; Based on the changes in the key assumptions, a record of viewpoint reversal events is generated.

[0006] Optionally, generating a viewpoint reversal event record based on changes in the key assumptions includes: Extract the first set of key assumptions from the first conflict report before the reversal point. The first set of key assumptions includes the first supply-demand gap assumption, the first inventory cycle assumption, or the first macroeconomic policy assumption. Extract a second set of key assumptions from the second conflict report after the reversal point. The second set of key assumptions includes a second supply-demand gap assumption, a second inventory cycle assumption, or a second macroeconomic policy assumption. By comparing the first set of key assumptions with the second set of key assumptions, the target assumptions that have changed are identified. Determine whether the direction of change of the target assumption is consistent with the reversal direction of the bullish / bearish direction label; If they match, the target assumption is marked as a reversal driver, and a reversal event record is generated that includes the reversal driver, the reversal time point, and the long / short direction labels before and after the reversal.

[0007] Optionally, after determining whether the direction of change of the target assumption is consistent with the reversal direction of the long / short direction label, the method further includes: If the direction of change of the target assumption is inconsistent with the reversal direction of the long / short direction label, then obtain the long / short direction labels of three consecutive reports before and after the reversal time. Determine whether the three consecutive reports show an N-type reversal pattern of first high, then low, then high again or first low, then high, then low again; If the N-type reversal pattern is presented, the external unforeseen factors cited in the middle report of the N-type reversal are extracted. These external unforeseen factors include geopolitical conflict events, extreme weather in the main production area, or temporary adjustments to margin ratios by the exchange. The external sudden factors are marked as temporary disturbance reversal factors, and the decay time window corresponding to the temporary disturbance reversal factors is generated; The external sudden factor is associated with the decay time window and stored in the reversal event record.

[0008] Optionally, after generating the decay time window corresponding to the temporary interference reversal factor, the method further includes: Obtain the real-time bullish / bearish directional labels of the five latest consecutive reports following the occurrence of the temporary disruptive reversal factor; Determine whether three or more of the five latest consecutive reports show a return to the original bullish / bearish direction label before the occurrence of the external emergency factor; If three or more consecutive data points revert to the original long / short direction labels, it is determined that the temporary interference reversal factor has decayed, and the decay time window is closed. If no more than three consecutive reports recover to the original bullish / bearish direction label, then check whether at least two of the five latest consecutive reports mention the aftereffects of the external sudden factor again in the supporting data; If at least two reports mention the aforementioned aftereffects, the decay time window is doubled, and monitoring continues for the next five reports.

[0009] Optionally, the step of fusing the core viewpoints and supporting data from the two adjacent reports to generate a consistent viewpoint summary includes: Extract the core viewpoint sentences from each of the two adjacent reports, and perform dependency parsing on each core viewpoint sentence to obtain the subject-verb-object structure of each core viewpoint sentence; Extract the futures trading target referred to by the subject, the trend direction verb indicated by the predicate, and the magnitude modifier carried by the object from the subject-verb-object structure. Compare whether the trend direction verbs in the two core viewpoint sentences belong to the same semantic direction; If they belong to the same semantic direction, then further compare whether there is an overlap in the numerical range expressed by the magnitude modifiers of the two core viewpoint sentences; If there are overlapping intervals, the minimum value of the overlapping interval is used as the lower limit and the maximum value is used as the upper limit to generate a fused viewpoint in the form of an amplitude interval. If there are no overlapping intervals, retain the magnitude modifiers of the two core viewpoint sentences respectively, and set the divergence magnitude indicator at the end of the consensus viewpoint summary; If they do not belong to the same semantic direction, then extract the publication time of the data cited in each of the two reports from the supporting data of each report; The core viewpoints of the report with updated data release time will be adopted as provisional viewpoints, while the core viewpoints of another report will be included as dissenting notes in the appendix of the consensus viewpoint summary.

[0010] Optionally, after generating the fused viewpoint in the form of an amplitude range, the method further includes: Obtain the historical forecast accuracy of the first publishing institution for the first report, and the historical forecast accuracy of the second publishing institution for the second report; Obtain the number of first data sources contained in the first supporting data of the first report, and the number of second data sources contained in the second supporting data of the second report; The historical prediction accuracy is multiplied by the number of data sources to obtain the first confidence score and the second confidence score, respectively. Determine whether the ratio between the first confidence score and the second confidence score exceeds a target multiple threshold; If the target multiple threshold is exceeded, the amplitude value corresponding to the report with the higher credibility score will be used as the single recommended value for the integrated viewpoint, and the amplitude contribution of the other report will be removed. If the target multiple threshold is not exceeded, the magnitude values ​​of the two reports are weighted and averaged according to the proportion of their respective credibility scores to the total score to generate a weighted magnitude range, and the magnitude range is used as the final output value of the integrated viewpoint. Obtain the maximum deviation between the weighted amplitude range and the original amplitude values ​​of the two reports. When the maximum deviation exceeds a preset deviation tolerance threshold, generate a deviation warning mark next to the fused viewpoint.

[0011] Optionally, determining whether the long / short direction labels of two adjacent reports in the time series report set have been reversed includes: Obtain the first bullish / bearish direction label from the previous report and the second bullish / bearish direction label from the next report in two adjacent reports; Convert the first bullish / bearish direction label and the second bullish / bearish direction label into numerical form respectively: convert the bullish label to +1, the bearish label to -1, and the neutral label to 0. Calculate the difference between the value of the second bullish / bearish direction label and the value of the first bullish / bearish direction label to obtain the direction change value; If the absolute value of the direction change is equal to 2, it is determined that a reversal has occurred; If the absolute value of the direction change value is not equal to 2, check whether there is a neutral label in the first long / short direction label or the second long / short direction label; If one label is neutral and the other label is non-neutral, then obtain the number of supporting data referenced in each of the two adjacent reports; The tags corresponding to reports with more supporting data will be considered valid tags. Determine whether the valid tag is the opposite of the tag corresponding to another report in terms of null and multiple semantics; If the opposite is true, then it is determined that a reversal has occurred.

[0012] Optionally, checking whether a neutral label exists in the first or second multi-short direction label includes: Read the tag name field of the first multi-space direction tag, and perform string matching between the tag name field and the preset neutral tag feature library; If the tag name field of the first multi-space direction tag successfully matches any target feature word, the output check result is that there is a neutral tag in the first multi-space direction tag, and the check of the second multi-space direction tag is stopped. If the tag name field of the first multi-short direction tag fails to match all the target feature words, then the tag name field of the second multi-short direction tag is read, and the tag name field of the second multi-short direction tag is matched with the target feature words again. If the label name field of the second multi-space direction label matches any target feature word, the output check result is that there is a neutral label in the second multi-space direction label; If the tag name field of the first multi-short direction tag and the second multi-short direction tag fails to match all the target feature words, then the context feature vectors of the first multi-short direction tag and the second multi-short direction tag are extracted respectively. The context feature vectors include the preceding verb and the following noun of the report paragraph in which the tag is located. Calculate the similarity between the context feature vector of each tag and the context feature vector of each target feature word in the neutral tag feature library, and take the maximum similarity as the neutral confidence of the tag; If the neutral confidence level of the first multi-short direction label is greater than the first confidence threshold, the output check result is that a neutral label exists in the first multi-short direction label; If the neutral confidence level of the first long / short direction label is not greater than the first confidence threshold, then determine whether the neutral confidence level of the second long / short direction label is greater than the second confidence threshold. If the neutral confidence level of the second long / short direction label is greater than the second confidence threshold, the output check result is that a neutral label exists in the second long / short direction label; If the neutral confidence level of the first multi-short direction label is not greater than the first confidence threshold and the neutral confidence level of the second multi-short direction label is not greater than the second confidence threshold, then the output check result is that there is no neutral label.

[0013] Optionally, after generating the viewpoint reversal event record based on the changes in the key assumptions, the method further includes: Retrieve all inversion event records for the currently active state; Extract reversal drivers or temporary disruptive reversal factors from each reversal event record; Based on the reversal driving factors or temporary interference reversal factors, multiple reversal event records are clustered to obtain at least one factor cluster; Count the number of reversal events contained in each factor cluster, and the number of futures contracts affected by each factor cluster; When the number of reversal events exceeds the first event threshold and the number of varieties exceeds the first variety threshold, the corresponding factor cluster is marked as a systematic reversal factor; Generate a cross-variety early warning signal containing the systemic reversal factors, and push the cross-variety early warning signal to the monitoring interface.

[0014] Secondly, this application also discloses a research report processing system for the futures market, which adopts the following technical solution: A research report processing system for the futures market includes: The data acquisition module is used to acquire multiple original investment research reports from different data sources; The information extraction module is used to extract the futures varieties, release time, core viewpoints, bullish / bearish direction tags and supporting data from each original investment research report; The sequence segmentation module is used to divide the multiple original investment research reports into a time series report set of the same futures product based on the futures product and the release time. The reversal judgment module is used to determine whether the long / short direction labels of two adjacent reports in the time series report set have been reversed; The viewpoint fusion module is used to merge the core viewpoints and supporting data of two adjacent reports to generate a consistent viewpoint summary when no reversal occurs. The conflict resolution module is used to extract two conflict reports before and after the reversal time when a reversal occurs, and to parse the key assumptions in the supporting data of the two conflict reports respectively. The event logging module is used to generate opinion reversal event logs based on changes in the key assumptions.

[0015] In summary, this application includes the following beneficial technical effects: By acquiring multiple original investment research reports from different data sources, the system automatically extracts the futures contracts, publication dates, core viewpoints, bullish / bearish direction tags, and supporting data from each report, effectively solving the problem of laborious and easily overlooked information when manually extracting report information. At the same time, by dividing multiple original investment research reports into time series reports of the same contract based on futures contracts and publication dates, the system automatically determines whether the bullish / bearish direction tags of adjacent reports have reversed. When a reversal occurs, the system extracts the key assumptions in the conflicting reports and generates a viewpoint reversal event record, solving the problem of low efficiency in manually identifying viewpoint reversal events. Attached Figure Description

[0016] Figure 1 This is a main flowchart of a futures research report processing method according to an embodiment of this application; Figure 2 This is a flowchart of steps S201 to S205; Figure 3 This is a flowchart of steps S301 to S305; Figure 4 This is a flowchart of steps S401 to S405; Figure 5 This is a flowchart of steps S501 to S508; Figure 6 This is a flowchart of steps S601 to S607; Figure 7 This is a flowchart of steps S701 to S709; Figure 8 This is a flowchart of steps S801 to S810; Figure 9 This is a flowchart of steps S901 to S906; Figure 10 This is a module diagram of a futures research report processing system according to an embodiment of this application.

[0017] Explanation of reference numerals in the attached figures: 1. Data acquisition module; 2. Information extraction module; 3. Sequence partitioning module; 4. Reversal judgment module; 5. Viewpoint fusion module; 6. Conflict resolution module; 7. Event logging module. Detailed Implementation

[0018] Firstly, this application discloses a method for processing investment research reports in the futures field.

[0019] Reference Figure 1 A method for processing investment research reports in the futures field, including steps S101 to S107: Step S101: Obtain multiple original investment research reports from different data sources.

[0020] Specifically, the data sources include official websites of futures exchanges, PDF or Word reports published by mainstream futures research institutions, HTML articles from financial information platforms, and structured report data provided by third-party data interfaces. The system obtains these original investment research reports through API interfaces, web crawlers, or file uploads. Each report includes information such as title, body text, publication date, and author / institution.

[0021] Step S102: Extract the futures contracts, release date, core viewpoints, bullish / bearish direction tags, and supporting data from each original investment research report.

[0022] Specifically, "futures product" refers to the specific futures contract underlying asset analyzed in the report, such as rebar, crude oil, and soybean meal. "Publication date" refers to the date the report was written or published, used to determine the chronological order of the reports. "Core viewpoint" refers to the report's main judgment and conclusion regarding the future price trend of the futures product, usually appearing in the summary or conclusion paragraph. "Bull / Bear" labels are simplified identifiers of the core viewpoint, categorized into bullish (predicting price increases), bearish (predicting price decreases), and neutral (believing price fluctuations or unclear direction). "Supporting data" refers to the numerical evidence used in the report to corroborate the core viewpoint, including inventory data, production data, price levels, and supply-demand imbalances. The system uses regular expression matching and natural language processing methods to extract the above information from the report.

[0023] Step S103: Based on the futures product and the release time, divide multiple original investment research reports into a time series report set for the same product.

[0024] Specifically, the system first groups all reports according to the futures contract, grouping reports analyzing the same contract into the same contract group. Then, it sorts the reports within each contract group in ascending order of publication time, forming a time series report set for that contract. For example, if a contract has report A (January), report B (February), and report C (March) on the timeline, they will be arranged in chronological order as [A, B, C].

[0025] Step S104: Determine whether the long / short direction labels of two adjacent reports in the time series report set have been reversed.

[0026] Specifically, a reversal refers to a change in the bullish or bearish direction label between two adjacent reports. The system sequentially retrieves two adjacent reports from the time series, reads the bullish or bearish direction label for each report, and compares whether the two labels are opposite. If the label changes from bullish to bearish or from bearish to bullish, a reversal is determined; if the labels are the same or contain a neutral label, further judgment is required.

[0027] Step S105: If no reversal occurs, the core viewpoints and supporting data of the two adjacent reports are merged to generate a consistent viewpoint summary.

[0028] Specifically, when the bullish / bearish direction labels of two adjacent reports are consistent or the system determines that a reversal is not possible, the system will merge the core viewpoints of the two reports, extract the commonly accepted price trend direction and range, and integrate the supporting data of the two reports to generate a concise summary of consistent viewpoints, allowing users to quickly understand the mainstream judgment of the current market.

[0029] Step S106: If a reversal occurs, extract the two conflict reports before and after the reversal time, and analyze the key assumptions in the supporting data of the two conflict reports respectively.

[0030] Specifically, when a reversal is determined, the reversal point refers to the time when the latter report was published. Two conflicting reports refer to the report before the reversal and the report after the reversal. Key assumptions are the presuppositional judgments in a report that support the core viewpoint; for example, "if the supply-demand gap widens, prices will rise" is a key assumption. The system identifies these key assumptions from the supporting data of each conflicting report, including supply-demand gap assumptions, inventory cycle assumptions, and macroeconomic policy assumptions.

[0031] Step S107: Generate a viewpoint reversal event record based on changes in key assumptions.

[0032] Specifically, the system compares the key assumptions in the two reports before and after the reversal to identify which assumptions have changed. If the direction of change in a key assumption aligns with the direction of the reversal label, the change is marked as a reversal driver. The system combines the reversal driver, the specific time of the reversal, and the bullish / bearish direction labels before and after the reversal into a complete reversal event record for subsequent analysis and early warning.

[0033] Reference Figure 2 In one embodiment of this example, generating a viewpoint reversal event record based on changes in key assumptions includes steps S201 to S205: Step S201: Extract the first set of key assumptions from the first conflict report before the reversal point. The first set of key assumptions includes the first supply-demand gap assumption, the first inventory cycle assumption, or the first macroeconomic policy assumption.

[0034] Specifically, the supply-demand gap assumption refers to the report's judgment on the difference between the supply and demand of futures contracts, such as "The supply-demand gap is expected to widen to 500,000 tons next quarter." The inventory cycle assumption refers to the report's judgment on the current stage of inventory (active replenishment, passive replenishment, active destocking, passive destocking), such as "Currently in the late stage of passive destocking." The macroeconomic policy assumption refers to the report's judgment on the direction of monetary policy, fiscal policy, or industrial policy, such as "The central bank is expected to cut interest rates by 25 basis points in the third quarter." The system locates and extracts these hypothetical statements from the supporting data area of ​​the report prior to the reversal point.

[0035] Step S202: Extract the second set of key assumptions from the second conflict report after the reversal point. The second set of key assumptions includes the second supply-demand gap assumption, the second inventory cycle assumption, or the second macroeconomic policy assumption.

[0036] Specifically, using the same method as in step S201, the corresponding key assumptions are extracted from the report after the reversal point. The key assumptions in the two reports may involve the same type of assumption (such as both being supply and demand gap assumptions), or they may involve different types of assumptions.

[0037] Step S203: Compare the first set of key assumptions with the second set of key assumptions to identify the target assumptions that have changed.

[0038] Specifically, the system compares pairs of assumptions of the same type in the first and second key sets of assumptions. For example, it compares the first supply-demand gap assumption with the second supply-demand gap assumption in terms of value or direction to determine whether there has been a change. If an assumption exists in the first report but disappears in the second report, or if its value or direction changes, then that assumption is marked as a changed target assumption.

[0039] Step S204: Determine whether the direction of change of the target assumption is consistent with the reversal direction of the long / short direction label.

[0040] Specifically, the direction of change refers to the increase or decrease, improvement or deterioration of the numerical value of the assumed condition. For example, if the supply-demand gap assumption changes from "gap widening" to "gap narrowing," then the direction of change is gap narrowing. The reversal direction refers to the change in the direction of the bullish / bearish label; for example, if it changes from bullish to bearish, then the reversal direction is bearish. The system judges whether the change in the assumed condition supports or explains the reversal of the label. For example, if the supply-demand gap narrowing (bearish) corresponds to the label changing from bullish to bearish, then it is judged as consistent.

[0041] Step S205: If consistent, mark the target assumption as a reversal driver and generate a reversal event record containing the reversal driver, the reversal time point, and the long / short direction labels before and after the reversal.

[0042] Specifically, when the direction of change aligns with the direction of reversal, the system considers the change in the assumption as the cause of the view reversal and marks it as a reversal driver. The system creates a reversal event record, which includes: the specific details of the reversal driver, the specific time point of the reversal (i.e., the release time of the subsequent report), the bullish / bearish direction label before the reversal, and the bullish / bearish direction label after the reversal. This record is stored in the database for subsequent querying and analysis.

[0043] Reference Figure 3 In one embodiment of this example, after determining whether the direction of change of the target assumption is consistent with the reversal direction of the bullish / bearish direction label, steps S301 to S305 are further included: Step S301: If the direction of change of the target assumption is inconsistent with the reversal direction of the long / short direction label, then obtain the long / short direction labels of the three consecutive reports before and after the reversal time.

[0044] Specifically, when the direction of change in the target assumptions cannot explain the label reversal, the system suspects that the reversal may be a temporary reversal caused by sudden external factors, rather than a trend reversal based on changes in fundamental assumptions. The system obtains the bullish / bearish direction labels from three consecutive reports before and after the reversal time: one before the reversal, one at the time of the reversal, and one after the reversal.

[0045] Step S302: Determine whether the three consecutive reports show an N-type reversal pattern of first long, then short, then long again or first short, then long, then short again.

[0046] Specifically, the N-shaped reversal pattern is a special reversal formation characterized by two directional changes in a short period, forming an N-shaped price action. The "bullish-bearish-bullish again" pattern refers to a report that is bullish first, then bearish second, and then bullish again third. The "bearish-bullish-bullish again" pattern refers to a report that is bearish first, then bullish second, and then bearish again third. This pattern typically indicates that the market quickly resumes its original trend after a short-term shock.

[0047] Step S303: If an N-type reversal pattern is presented, extract the external unforeseen factors cited in the middle report of the N-type reversal. External unforeseen factors include geopolitical conflict events, extreme weather in the main production area, or temporary adjustments to margin ratios by the exchange.

[0048] Specifically, the middle report in an N-shaped reversal pattern is usually the key report for the reversal to occur. The system analyzes the content of this report and extracts the external unforeseen factors cited within. Geopolitical conflicts refer to political emergencies such as war, sanctions, and trade frictions. Extreme weather in major producing areas refers to weather events such as floods, droughts, and hurricanes that affect agricultural or energy production. Temporary adjustments to margin ratios by exchanges refer to the actions of futures exchanges in temporarily increasing or decreasing trading margins to control risk.

[0049] Step S304: Mark the external sudden factors as temporary disturbance reversal factors and generate the decay time window corresponding to the temporary disturbance reversal factors.

[0050] Specifically, the impact of external unforeseen factors is usually short-term and diminishes over time. The system marks such external unforeseen factors as temporary disturbance reversal factors and generates a decay time window based on the historical duration of the factor's impact. For example, the decay time window for geopolitical conflict events is typically set to 7 to 14 days, for extreme weather events to 5 to 10 days, and for margin adjustments to 3 to 7 days.

[0051] Step S305: Associate the external sudden factor with the decay time window and store it in the inversion event record.

[0052] Specifically, the system stores the identified external sudden factors and their corresponding decay time windows as additional information in the reversal event log. This allows subsequent processing to determine whether the temporary disturbance reversal factor is still effective or has completely decayed based on the decay time window.

[0053] Reference Figure 4 In one embodiment of this invention, after generating the decay time window corresponding to the temporary interference reversal factor, steps S401 to S405 are further included: Step S401: Obtain the real-time bullish / bearish direction labels of the five latest reports following the occurrence of the temporary disturbance reversal factor.

[0054] Specifically, to monitor the decay of temporary disruptive reversal factors, the system continuously acquires the bullish / bearish directional labels from the five latest reports following the occurrence of a temporary disruptive reversal factor. These five reports are ordered by publication time, and each report carries its corresponding bullish / bearish directional label.

[0055] Step S402: Determine whether three or more of the five latest reports show a return to the original bullish / bearish direction label before the occurrence of the external unforeseen factor.

[0056] Specifically, the original bullish / bearish direction label refers to the prevailing direction label before the occurrence of unforeseen external factors and when the market was undisturbed. For example, if reports before the occurrence of unforeseen external factors were generally bullish, then the original direction would be bullish. The system checks whether three or more consecutive reports in the subsequent five reports have the same bullish / bearish direction label as the original direction label.

[0057] Step S403: If three or more consecutive labels return to the original long / short direction, it is determined that the temporary interference reversal factor has decayed, and the decay time window is closed.

[0058] Specifically, when three or more consecutive reports return to the original direction, it indicates that the impact of the external unexpected factor has subsided, and the market has returned to its original trend. The system updates the status of this temporary disruptive reversal factor to "decayed" and closes the corresponding decay time window, ceasing continuous monitoring of the factor.

[0059] Step S404: If no more than three consecutive reports return to the original bullish / bearish direction label, check if at least two of the five latest reports mention the aftermath of external unforeseen factors again in their supporting data.

[0060] Specifically, if three or more of the subsequent five reports fail to restore the original direction, it indicates that the impact of the external emergency is still ongoing. The system further checks whether at least two of these five reports mention the subsequent impact or aftermath of the external emergency again in their supporting data areas, such as statements like "the aftermath of the geopolitical conflict is still ongoing" or "the lag effect of extreme weather is beginning to appear."

[0061] Step S405: If at least two reports mention the aftereffects, double the decay time window and continue monitoring for the next five reports.

[0062] Specifically, when at least two reports mention the aftereffects, it indicates that the impact of the external emergency has lasted longer than expected. The system doubles the current decay time window, for example, from 7 days to 14 days, and continues to monitor the next five reports, repeating steps S401 to S405 until the factor has decayed or the maximum number of monitoring attempts has been reached.

[0063] Reference Figure 5 In one embodiment of this example, the core viewpoints and supporting data of two adjacent reports are fused to generate a consistent viewpoint summary, including steps S501 to S508: Step S501: Extract the core viewpoint sentences from each of the two adjacent reports, and perform dependency parsing on each core viewpoint sentence to obtain the subject-verb-object structure of each core viewpoint sentence.

[0064] Specifically, dependency parsing is a natural language processing technique used to analyze the dependency relationships between words in a sentence. For example, in the sentence "The price of rebar is expected to rise by 200 points," "rebar price" is the subject, "rise" is the predicate, and "200 points" is the object. The system uses dependency parsing to identify the subject, predicate, and object components of each core viewpoint sentence.

[0065] Step S502: Extract the futures trading target referred to by the subject, the trend direction verb indicated by the predicate, and the magnitude modifier carried by the object from the subject-verb-object structure.

[0066] Specifically, the subject refers to the underlying asset of the futures transaction, such as "rebar," "crude oil," or "soybean meal." The predicate indicates the direction of the trend, using verbs like "rise," "fall," "rebound," and "retreat," to determine the direction of price movement. The object carries modifiers of magnitude, including specific numerical values ​​such as "200 points," "5%," or "100 yuan / ton," to quantify the degree of price change.

[0067] Step S503: Compare whether the trend direction verbs of the two core viewpoint sentences belong to the same semantic direction.

[0068] Specifically, the system determines whether the trend direction verbs in the two reports express the same direction. For example, "rise" and "rebound" both indicate a bullish direction, while "fall" and "retreat" both indicate a bearish direction. "Rise" and "fall" represent different directions. If the direction verbs in the two reports share the same semantic direction, the system proceeds to step S504; otherwise, it proceeds to step S507.

[0069] Step S504: If they belong to the same semantic direction, then further compare whether there is an overlap in the numerical range expressed by the magnitude modifiers of the two core viewpoint sentences.

[0070] Specifically, when two reports are aligned in direction, the system further compares whether their predicted magnitude ranges overlap. For example, if report A predicts an increase of 100-200 points and report B predicts an increase of 150-250 points, then the overlapping range is 150-200 points. If report A predicts an increase of 100-150 points and report B predicts an increase of 200-250 points, then there is no overlapping range.

[0071] Step S505: If there are overlapping intervals, the minimum value of the overlapping interval is used as the lower limit and the maximum value is used as the upper limit to generate a fused viewpoint in the form of an amplitude interval.

[0072] Specifically, when overlapping intervals exist, the system uses the lower bound of the overlapping interval as the lower limit of the convergent view's magnitude and the upper bound as the upper limit, generating a range-based prediction. For example, if the overlapping interval is 150-200 points, the convergent view would be "expected to rise by 150-200 points." This range-based format reflects the consensus range between different reports more accurately than a single numerical value.

[0073] Step S506: If there is no overlapping interval, retain the magnitude modifiers of the two core viewpoint sentences respectively, and set the divergence magnitude indicator at the end of the consensus viewpoint summary.

[0074] Specifically, when two reports share the same direction but their magnitude predictions do not overlap at all, it indicates a discrepancy between the two institutions' forecasts. The system retains both different magnitude predictions in the consensus summary and adds a discrepancy magnitude indicator at the end of the summary to alert users to significant differences in the magnitude predictions for that product.

[0075] Step S507: If they do not belong to the same semantic direction, extract the publication time of the data cited in each of the two reports from the supporting data.

[0076] Specifically, when two reports have different directions, the system considers them to have a fundamental disagreement. In this case, the system searches for the publication date cited in the supporting data of each report. The data publication date could be the date of official data released by the National Bureau of Statistics, the date of transaction data released by the exchange, or the cutoff date of data from a third-party survey.

[0077] Step S508: Take the core viewpoints of the report with updated data release time as the provisional adopted viewpoints, and store the core viewpoints of the other report as dissenting notes in the appendix of the consensus viewpoint summary.

[0078] Specifically, the system compares the timestamps of the data cited in the two reports and selects the core viewpoint from the report with the more recently updated data as the provisional adopted viewpoint, since more recent data is generally more valuable. The core viewpoint of the other report is not directly displayed in the abstract body, but is stored as a dissenting note in the appendix for user reference.

[0079] Reference Figure 6 In one embodiment of this example, after generating the fused viewpoint in the form of amplitude intervals, steps S601 to S607 are further included: Step S601: Obtain the historical forecast accuracy of the first issuing institution of the first report and the historical forecast accuracy of the second issuing institution of the second report.

[0080] Specifically, historical forecast accuracy refers to the proportion of forecasts issued by an institution over a past period (usually one year) that match actual market performance. For example, if an institution has issued 100 forecasts in the past, and 70 of them were correct, then its historical forecast accuracy is 70%. The system queries the historical database for the accuracy records of each institution.

[0081] Step S602: Obtain the number of first data sources contained in the first supporting data of the first report and the number of second data sources contained in the second supporting data of the second report.

[0082] Specifically, the number of data sources refers to the number of independent data sources cited in each report. For example, if a report cites three sources simultaneously: exchange inventory data, customs import and export data, and industry research data, then the number of data sources is 3. A higher number of data sources generally indicates a higher level of credibility for the report.

[0083] Step S603: Multiply the historical prediction accuracy by the number of data sources to obtain the first confidence score and the second confidence score.

[0084] Specifically, the system multiplies each report's historical prediction accuracy (expressed as a decimal) by the number of its data sources to obtain the report's credibility score. For example, a report with a historical prediction accuracy of 0.7 and 3 data sources would have a credibility score of 2.1. This calculation method considers both the institution's predictive ability and the sufficiency of the report's evidence.

[0085] Step S604: Determine whether the ratio between the first confidence score and the second confidence score exceeds the target multiple threshold.

[0086] Specifically, the target multiple threshold is set to 3. The system calculates the ratio of the larger confidence score to the smaller confidence score and determines whether this ratio is greater than 3. For example, if the first confidence score is 2.1 and the second confidence score is 0.5, the ratio is 4.2, which exceeds the threshold of 3.

[0087] Step S605: If the target multiple threshold is exceeded, the amplitude value corresponding to the report with the higher credibility score is used as the single recommended value of the integrated viewpoint, and the amplitude contribution of the other report is removed.

[0088] Specifically, when the credibility of one report is significantly higher than that of another (more than 3 times), the system considers the magnitude prediction of the high-credibility report to be more trustworthy and directly uses the magnitude end value of that report as the final output of the fusion view. The magnitude prediction of the low-credibility report is not included in the fusion.

[0089] Step S606: If the target multiple threshold is not exceeded, the amplitude values ​​of the two reports are weighted and averaged according to the proportion of their respective credibility scores to the total score to generate a weighted amplitude range, and the amplitude range is used as the final output value of the fused viewpoints.

[0090] Specifically, when the credibility scores of two reports are similar (the ratio does not exceed 3), the system uses a weighted average for fusion. For example, report A has a credibility score of 2.1, with a lower limit of 100 and an upper limit of 150; report B has a credibility score of 1.5, with a lower limit of 120 and an upper limit of 160. The total score is 3.6. The weight of report A is 2.1 / 3.6 ≈ 0.583, and the weight of report B is 1.5 / 3.6 ≈ 0.417. The lower limit after weighting is 100 × 0.583 + 120 × 0.417 ≈ 108, and the upper limit after weighting is 150 × 0.583 + 160 × 0.417 ≈ 154.

[0091] Step S607: Obtain the maximum deviation between the weighted amplitude range and the original amplitude values ​​of the two reports. When the maximum deviation exceeds the preset deviation tolerance threshold, generate a deviation warning mark next to the fused viewpoint.

[0092] Specifically, the system calculates the deviation between the weighted fusion amplitude range and the original amplitude values ​​of each report, and takes the maximum value. For example, if the fusion result is 108-154, the maximum deviation from report A's 100-150 is 4 (from 150 to 154), and the maximum deviation from report B's 120-160 is 12 (from 120 to 108). The preset deviation tolerance threshold is set to 10. Since 12 is greater than 10, the system generates a deviation warning mark next to the fusion viewpoint, indicating to the user that the fusion result differs significantly from a certain original report.

[0093] Reference Figure 7 In one embodiment of this invention, determining whether the long / short direction labels of two adjacent reports in the time series report set are reversed includes steps S701 to S709: Step S701: Obtain the first bullish / bearish direction label of the previous report and the second bullish / bearish direction label of the next report from two adjacent reports.

[0094] Specifically, two adjacent reports are taken from the time series report set in sequence. The long / short direction label of the first report is read as the first long / short direction label, and the long / short direction label of the second report is read as the second long / short direction label.

[0095] Step S702: Convert the first bullish / bearish direction label and the second bullish / bearish direction label into numerical form respectively, convert the bullish label to +1, the bearish label to -1, and the neutral label to 0.

[0096] Specifically, to facilitate the calculation of differences between labels, the system maps three types of labels to numerical values: +1 for bullish, -1 for bearish, and 0 for neutral. This mapping relationship ensures that the difference between labels with opposite directions (+1 and -1) is 2 or -2, with an absolute value equal to 2.

[0097] Step S703: Calculate the difference between the value of the second bullish / bearish direction label and the value of the first bullish / bearish direction label to obtain the direction change value.

[0098] Specifically, the direction change value = second label value - first label value. For example, changing from bullish (+1) to bearish (-1), the direction change value is -2, and the absolute value is 2. Changing from bullish (+1) to neutral (0), the direction change value is -1, and the absolute value is 1. Changing from bearish (-1) to bullish (+1), the direction change value is +2, and the absolute value is 2.

[0099] Step S704: If the absolute value of the direction change is equal to 2, it is determined that a reversal has occurred.

[0100] Specifically, when the absolute value of the direction change is equal to 2, it indicates that the label has changed from bullish to bearish or from bearish to bullish without the transition of a neutral label. The system directly determines that a reversal has occurred and proceeds to step S106.

[0101] Step S705: If the absolute value of the direction change value is not equal to 2, check whether there is a neutral label in the first long / short direction label or the second long / short direction label.

[0102] Specifically, when the absolute value of the direction change is 0 or 1, it indicates that the two labels are the same or one of them is a neutral label. The system further checks whether a neutral label (with a value of 0) exists between the two labels, because the presence of a neutral label may lead to a hidden reversal signal.

[0103] Step S706: If one of them is a neutral label and the other is a non-neutral label, then obtain the number of supporting data referenced in each of the two adjacent reports.

[0104] Specifically, when one report is labeled neutrally (e.g., "volatile") and another is labeled non-neutrally (e.g., "bullish" or "bearish"), the system needs to determine whether the neutral label implies the true direction. The system obtains the number of supporting data entries cited in each report. Supporting data refers to numerical evidence used in the report to corroborate the viewpoint; more data usually indicates a more reliable viewpoint.

[0105] Step S707: The labels corresponding to reports that cite more supporting data are considered valid labels.

[0106] Specifically, the system compares the amount of data supporting the two reports. More data means more sufficient argumentation and higher credibility. Therefore, the label of the report with more data is selected as the valid label.

[0107] Step S708: Determine whether the valid label and the label corresponding to another report are opposite in terms of empty and multiple semantics.

[0108] Specifically, the system determines whether a valid label and a label in another report are opposite in terms of whether they are bullish or bearish. For example, if a valid label is bullish and another label is bearish, then it is considered opposite. If a valid label is bullish and another label is neutral, then it is not opposite.

[0109] Step S709: If the opposite is true, then it is determined that a reversal has occurred.

[0110] Specifically, when a valid label is opposite to the label in another report in terms of null and multiple semantics, it indicates that the neutral label actually masks the change in direction, and the system judges it as a reversal.

[0111] Reference Figure 8In one embodiment of this example, checking whether a neutral label exists in the first or second multi-space direction label includes steps S801 to S810: Step S801: Read the tag name field of the first long / short direction tag and perform string matching between the tag name field and the preset neutral tag feature library.

[0112] Specifically, the tag name field contains the original text of the bullish / bearish directional tags extracted during report parsing, such as "consolidation," "neutral," and "unclear direction." The system pre-sets a neutral tag feature library containing six target feature words: "consolidation," "wait and see," "neutral," "unclear direction," and "cautious." The system performs string matching between the tag name field of the first bullish / bearish directional tag and these six feature words one by one.

[0113] Step S802: If the label name field of the first multi-space direction label successfully matches any target feature word, the output check result is that there is a neutral label in the first multi-space direction label, and the check of the second multi-space direction label is stopped.

[0114] Specifically, if the label name field of the first long / short direction label contains feature words such as "oscillation" or "neutral," then the first long / short direction label is determined to be a neutral label. The system outputs this check result and stops subsequent checks, no longer matching the second long / short direction label.

[0115] Step S803: If the tag name field of the first multi-short direction tag fails to match all target feature words, then read the tag name field of the second multi-short direction tag and perform string matching between the tag name field of the second multi-short direction tag and the target feature words again.

[0116] Specifically, when the tag name field of the first long / short direction tag does not contain any of the six feature words, the system determines that the first long / short direction tag is not a neutral tag. At this time, the system continues to read the tag name field of the second long / short direction tag and uses the same method to perform string matching with each of the six target feature words in the neutral tag feature library.

[0117] Step S804: If the label name field of the second multi-space direction label matches any target feature word, the output check result is that there is a neutral label in the second multi-space direction label.

[0118] Specifically, if the tag name field of the second long / short direction tag contains any one of the six target feature words, then the second long / short direction tag is determined to be a neutral tag, and the system outputs the check result as the existence of a neutral tag in the second long / short direction tag.

[0119] Step S805: If the tag name field of the first and second multi-short direction tags fails to match all target feature words, then extract the context feature vectors of the first and second multi-short direction tags respectively. The context feature vectors include the preceding verb and following noun of the report paragraph in which the tag is located.

[0120] Specifically, when neither report's tag name field contains any neutral feature words, the system enters the semantic similarity judgment stage. The system locates the paragraph in the report containing each tag, extracts the two verbs immediately preceding the tag name as pre-verbs (e.g., "expected" and "possible" before the tag "oscillation"), and simultaneously extracts the two nouns immediately following the tag name as post-nouns (e.g., "market trend" and "price trend" after the tag "oscillation"). The pre-verbs and post-nouns together constitute the contextual feature vector of that tag.

[0121] Step S806: Calculate the similarity between the context feature vector of each tag and the context feature vector of each target feature word in the neutral tag feature library, and take the maximum similarity as the neutral confidence of the tag.

[0122] Specifically, the system pre-constructs a standard contextual feature vector for each target feature word (such as "oscillation") in the neutral label feature library. This vector is statistically derived based on a large number of report paragraphs already labeled as neutral. The system calculates the cosine similarity between the contextual feature vector of the first multi-short direction label and the standard contextual feature vector of each target feature word, and takes the maximum similarity as the neutral confidence score of the first multi-short direction label. The neutral confidence score of the second multi-short direction label is calculated in the same way. The neutral confidence score ranges from 0 to 1, with a higher value indicating that the label is semantically closer to a neutral label.

[0123] Step S807: If the neutral confidence level of the first long / short direction label is greater than the first confidence threshold, then the check result is output as: there is a neutral label in the first long / short direction label.

[0124] Specifically, the first confidence threshold is preset to 0.7. If the neutral confidence of the first multi-space direction label is greater than 0.7, it means that although the text form of the label is not a standard neutral word, its semantic expression is close to neutral. The system judges it as an implicit neutral label and outputs the check result as a neutral label exists in the first multi-space direction label.

[0125] Step S808: If the neutral confidence level of the first long / short direction label is not greater than the first confidence threshold, then determine whether the neutral confidence level of the second long / short direction label is greater than the second confidence threshold.

[0126] Specifically, when the neutral confidence level of the first long / short direction label is less than or equal to 0.7, the system continues to judge the second long / short direction label. The second confidence level threshold is also preset to 0.7.

[0127] Step S809: If the neutral confidence level of the second long / short direction label is greater than the second confidence threshold, then the output check result is that a neutral label exists in the second long / short direction label.

[0128] Specifically, if the neutral confidence level of the second long / short direction label is greater than 0.7, the system determines that the second long / short direction label is an implicit neutral label and outputs the check result that a neutral label exists in the second long / short direction label.

[0129] Step S810: If the neutral confidence of the first long / short direction label is not greater than the first confidence threshold and the neutral confidence of the second long / short direction label is not greater than the second confidence threshold, then the output check result is that there is no neutral label.

[0130] Specifically, when the neutral confidence scores of both labels are less than or equal to 0.7, it indicates that neither report's long / short direction labels possesses neutral semantic features, and the system outputs a check result indicating that no neutral labels exist. This result will be returned to step S705 for subsequent inversion determination logic.

[0131] Reference Figure 9 In one embodiment of this example, after generating a viewpoint reversal event record based on changes in key assumptions, steps S901 to S906 are further included: Step S901: Obtain all inversion event records for the currently active state.

[0132] Specifically, an "active" status refers to a reversal event record that is in effect and has not yet been closed or marked as processed. The system queries the database for all reversal event records with an "active" status. These records may come from reversal events occurring in different futures contracts and at different times. Each reversal event record must contain at least the reversal driving factor or temporary interfering reversal factor, the reversal time, and the bullish / bearish direction labels before and after the reversal.

[0133] Step S902: Extract the reversal driving factor or temporary interference reversal factor from each reversal event record.

[0134] Specifically, the system iterates through each active reversal event record, extracting either the reversal driver or the temporary disruptive reversal factor. The reversal driver refers to a change in the fundamental assumptions that causes a reversal of opinion, such as "the supply-demand gap shifting from widening to narrowing" or "the inventory cycle shifting from destocking to inventory accumulation." The temporary disruptive reversal factor refers to sudden external factors that cause a temporary reversal, such as "geopolitical conflict events" or "extreme weather in major producing areas." Each reversal event record must contain at least one of these two types of factors.

[0135] Step S903: Cluster multiple reversal event records based on reversal driving factors or temporary interference reversal factors to obtain at least one factor cluster.

[0136] Specifically, the system clusters all extracted factors based on semantic similarity. For example, all factors containing expressions related to "supply and demand gap" are grouped into one cluster, all factors containing expressions related to "inventory cycle" are grouped into another cluster, and all factors containing expressions related to "geopolitical conflict" are grouped into yet another cluster. After clustering, each cluster is called a factor cluster, representing a set of inversion events with the same or similar driving logic.

[0137] Step S904: Count the number of reversal events contained in each factor cluster, and the number of futures contracts affected by each factor cluster.

[0138] Specifically, for each factor cluster, the system counts two indicators: the first indicator is the total number of reversal events contained in the factor cluster, for example, the "supply and demand gap" factor cluster contains 15 reversal event records; the second indicator is the number of duplicate futures contracts involved in all reversal events in the factor cluster, for example, these 15 reversal events involve three contracts: rebar, hot-rolled coil, and iron ore, so the number of contracts is 3.

[0139] Step S905: When the number of reversal events exceeds the first event threshold and the number of varieties exceeds the first variety threshold, the corresponding factor cluster is marked as a systematic reversal factor.

[0140] Specifically, the first event threshold is preset to 10, and the first commodity threshold is preset to 3. When the number of reversal events for a factor cluster is greater than 10, and the number of futures commodities affected by the factor cluster is greater than 3, it indicates that the factor is not an independent phenomenon of a single commodity, but a systemic factor affecting multiple commodities. The system marks the factor corresponding to the factor cluster (such as "supply and demand gap") as a systemic reversal factor, indicating that a change in this factor may trigger a collective reversal of multiple futures commodities.

[0141] Step S906: Generate a cross-variety early warning signal containing systemic reversal factors and push the cross-variety early warning signal to the monitoring interface.

[0142] Specifically, the system generates a cross-commodity early warning signal based on the cluster of factors marked as systemic reversal factors. This early warning signal includes at least the following information: the name of the systemic reversal factor (e.g., "narrowing supply-demand gap"), the affected timeframe, a list of involved futures contracts (e.g., "rebar, hot-rolled coil, iron ore"), and the warning level (assessed based on the number of reversal events and the number of contracts). The system pushes this early warning signal to a prominent position on the monitoring interface, allowing investment research personnel to promptly pay attention to potential cross-commodity trend changes.

[0143] Secondly, this application also discloses a research report processing system for the futures market.

[0144] Reference Figure 10 A research report processing system for the futures market, comprising: The data acquisition module is used to acquire multiple original investment research reports from different data sources; The information extraction module is used to extract the futures varieties, release time, core viewpoints, bullish / bearish direction tags and supporting data from each original investment research report; The sequence segmentation module is used to divide multiple original investment research reports into a time series report set for the same futures product based on the futures product and the release time. The reversal judgment module is used to determine whether the long / short direction labels of two adjacent reports in a time series report set have been reversed; The viewpoint fusion module is used to merge the core viewpoints and supporting data of two adjacent reports to generate a consistent viewpoint summary when no reversal occurs. The conflict resolution module is used to extract two conflict reports before and after the reversal point when a reversal occurs, and to analyze the key assumptions in the supporting data of the two conflict reports respectively. The event logging module is used to generate opinion reversal event logs based on changes in key assumptions.

[0145] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for processing investment research reports in the futures market, characterized in that, include: Obtain multiple original investment research reports from different data sources; Extract the futures contracts, release date, core viewpoints, bullish / bearish directional tags, and supporting data from each original investment research report; Based on the futures product and the publication time, the multiple original investment research reports are divided into a time series report set for the same product; Determine whether the long / short direction labels of two adjacent reports in the time series report set have been reversed; If no reversal occurs, the core viewpoints and supporting data of the two adjacent reports will be merged to generate a consistent viewpoint summary; If a reversal occurs, extract the two conflict reports before and after the reversal time, and analyze the key assumptions in the supporting data of the two conflict reports respectively; Based on the changes in the key assumptions, a record of viewpoint reversal events is generated.

2. The method for processing investment research reports in the futures field according to claim 1, characterized in that, The generation of opinion reversal event records based on changes in the key assumptions includes: Extract the first set of key assumptions from the first conflict report before the reversal point. The first set of key assumptions includes the first supply-demand gap assumption, the first inventory cycle assumption, or the first macroeconomic policy assumption. Extract a second set of key assumptions from the second conflict report after the reversal point. The second set of key assumptions includes a second supply-demand gap assumption, a second inventory cycle assumption, or a second macroeconomic policy assumption. By comparing the first set of key assumptions with the second set of key assumptions, the target assumptions that have changed are identified. Determine whether the direction of change of the target assumption is consistent with the reversal direction of the bullish / bearish direction label; If they match, the target assumption is marked as a reversal driver, and a reversal event record is generated that includes the reversal driver, the reversal time point, and the long / short direction labels before and after the reversal.

3. The method for processing investment research reports in the futures field according to claim 2, characterized in that, After determining whether the direction of change of the target assumption is consistent with the reversal direction of the long / short direction label, the method further includes: If the direction of change of the target assumption is inconsistent with the reversal direction of the long / short direction label, then obtain the long / short direction labels of three consecutive reports before and after the reversal time. Determine whether the three consecutive reports show an N-type reversal pattern of first high, then low, then high again or first low, then high, then low again; If the N-type reversal pattern is presented, the external unforeseen factors cited in the middle report of the N-type reversal are extracted. These external unforeseen factors include geopolitical conflict events, extreme weather in the main production area, or temporary adjustments to margin ratios by the exchange. The external sudden factors are marked as temporary disturbance reversal factors, and the decay time window corresponding to the temporary disturbance reversal factors is generated; The external sudden factor is associated with the decay time window and stored in the reversal event record.

4. The method for processing investment research reports in the futures field according to claim 3, characterized in that, After generating the decay time window corresponding to the temporary interference reversal factor, the method further includes: Obtain the real-time bullish / bearish directional labels of the five latest consecutive reports following the occurrence of the temporary disruptive reversal factor; Determine whether three or more of the five latest consecutive reports show a return to the original bullish / bearish direction label before the occurrence of the external emergency factor; If three or more consecutive data points revert to the original long / short direction labels, it is determined that the temporary interference reversal factor has decayed, and the decay time window is closed. If no more than three consecutive reports recover to the original bullish / bearish direction label, then check whether at least two of the five latest consecutive reports mention the aftereffects of the external sudden factor again in the supporting data; If at least two reports mention the aforementioned aftereffects, the decay time window is doubled, and monitoring continues for the next five reports.

5. The method for processing investment research reports in the futures field according to claim 1, characterized in that, The step of fusing the core viewpoints and supporting data from the two adjacent reports to generate a consistent viewpoint summary includes: Extract the core viewpoint sentences from each of the two adjacent reports, and perform dependency parsing on each core viewpoint sentence to obtain the subject-verb-object structure of each core viewpoint sentence; Extract the futures trading target referred to by the subject, the trend direction verb indicated by the predicate, and the magnitude modifier carried by the object from the subject-verb-object structure. Compare whether the trend direction verbs in the two core viewpoint sentences belong to the same semantic direction; If they belong to the same semantic direction, then further compare whether there is an overlap in the numerical range expressed by the magnitude modifiers of the two core viewpoint sentences; If there are overlapping intervals, the minimum value of the overlapping interval is used as the lower limit and the maximum value is used as the upper limit to generate a fused viewpoint in the form of an amplitude interval. If there are no overlapping intervals, retain the magnitude modifiers of the two core viewpoint sentences respectively, and set the divergence magnitude indicator at the end of the consensus viewpoint summary; If they do not belong to the same semantic direction, then extract the publication time of the data cited in each of the two reports from the supporting data of each report; The core viewpoints of the report with updated data release time will be adopted as provisional viewpoints, while the core viewpoints of another report will be included as dissenting notes in the appendix of the consensus viewpoint summary.

6. A method for processing investment research reports in the futures field according to claim 5, characterized in that, Following the generation of the fused viewpoint in the form of an amplitude range, the following is also included: Obtain the historical forecast accuracy of the first publishing institution for the first report, and the historical forecast accuracy of the second publishing institution for the second report; Obtain the number of first data sources contained in the first supporting data of the first report, and the number of second data sources contained in the second supporting data of the second report; The historical prediction accuracy is multiplied by the number of data sources to obtain the first confidence score and the second confidence score, respectively. Determine whether the ratio between the first confidence score and the second confidence score exceeds a target multiple threshold; If the target multiple threshold is exceeded, the amplitude value corresponding to the report with the higher credibility score will be used as the single recommended value for the integrated viewpoint, and the amplitude contribution of the other report will be removed. If the target multiple threshold is not exceeded, the magnitude values ​​of the two reports are weighted and averaged according to the proportion of their respective credibility scores to the total score to generate a weighted magnitude range, and the magnitude range is used as the final output value of the integrated viewpoint. Obtain the maximum deviation between the weighted amplitude range and the original amplitude values ​​of the two reports. When the maximum deviation exceeds a preset deviation tolerance threshold, generate a deviation warning mark next to the fused viewpoint.

7. A method for processing investment research reports in the futures field according to claim 1, characterized in that, The step of determining whether the bullish / bearish directional labels of two adjacent reports in the time series report set have been reversed includes: Obtain the first bullish / bearish direction label from the previous report and the second bullish / bearish direction label from the next report in two adjacent reports; Convert the first bullish / bearish direction label and the second bullish / bearish direction label into numerical form respectively: convert the bullish label to +1, the bearish label to -1, and the neutral label to 0. Calculate the difference between the value of the second bullish / bearish direction label and the value of the first bullish / bearish direction label to obtain the direction change value; If the absolute value of the direction change is equal to 2, it is determined that a reversal has occurred; If the absolute value of the direction change value is not equal to 2, check whether there is a neutral label in the first long / short direction label or the second long / short direction label; If one label is neutral and the other label is non-neutral, then obtain the number of supporting data referenced in each of the two adjacent reports; The tags corresponding to reports with more supporting data will be considered valid tags. Determine whether the valid tag is the opposite of the tag corresponding to another report in terms of null and multiple semantics; If the opposite is true, then it is determined that a reversal has occurred.

8. A method for processing investment research reports in the futures field according to claim 7, characterized in that, The step of checking whether a neutral label exists in the first or second multi-short direction label includes: Read the tag name field of the first multi-space direction tag, and perform string matching between the tag name field and the preset neutral tag feature library; If the tag name field of the first multi-space direction tag successfully matches any target feature word, the output check result is that there is a neutral tag in the first multi-space direction tag, and the check of the second multi-space direction tag is stopped. If the tag name field of the first multi-short direction tag fails to match all the target feature words, then the tag name field of the second multi-short direction tag is read, and the tag name field of the second multi-short direction tag is matched with the target feature words again. If the label name field of the second multi-space direction label matches any target feature word, the output check result is that there is a neutral label in the second multi-space direction label; If the tag name field of the first multi-short direction tag and the second multi-short direction tag fails to match all the target feature words, then the context feature vectors of the first multi-short direction tag and the second multi-short direction tag are extracted respectively. The context feature vectors include the preceding verb and the following noun of the report paragraph in which the tag is located. Calculate the similarity between the context feature vector of each tag and the context feature vector of each target feature word in the neutral tag feature library, and take the maximum similarity as the neutral confidence of the tag; If the neutral confidence level of the first multi-short direction label is greater than the first confidence threshold, the output check result is that a neutral label exists in the first multi-short direction label; If the neutral confidence level of the first long / short direction label is not greater than the first confidence threshold, then determine whether the neutral confidence level of the second long / short direction label is greater than the second confidence threshold. If the neutral confidence level of the second long / short direction label is greater than the second confidence threshold, the output check result is that a neutral label exists in the second long / short direction label; If the neutral confidence level of the first multi-short direction label is not greater than the first confidence threshold and the neutral confidence level of the second multi-short direction label is not greater than the second confidence threshold, then the output check result is that there is no neutral label.

9. A method for processing investment research reports in the futures field according to claim 1, characterized in that, After generating the viewpoint reversal event record based on the changes in the key assumptions, the process further includes: Retrieve all inversion event records for the currently active state; Extract reversal drivers or temporary disruptive reversal factors from each reversal event record; Based on the reversal driving factors or temporary interference reversal factors, multiple reversal event records are clustered to obtain at least one factor cluster; Count the number of reversal events contained in each factor cluster, and the number of futures contracts affected by each factor cluster; When the number of reversal events exceeds the first event threshold and the number of varieties exceeds the first variety threshold, the corresponding factor cluster is marked as a systematic reversal factor; Generate a cross-variety early warning signal containing the systemic reversal factors, and push the cross-variety early warning signal to the monitoring interface.

10. A research report processing system for the futures market, characterized in that, include: The data acquisition module is used to acquire multiple original investment research reports from different data sources; The information extraction module is used to extract the futures varieties, release time, core viewpoints, bullish / bearish direction tags and supporting data from each original investment research report; The sequence segmentation module is used to divide the multiple original investment research reports into a time series report set of the same futures product based on the futures product and the release time. The reversal judgment module is used to determine whether the long / short direction labels of two adjacent reports in the time series report set have been reversed; The viewpoint fusion module is used to merge the core viewpoints and supporting data of two adjacent reports to generate a consistent viewpoint summary when no reversal occurs. The conflict resolution module is used to extract two conflict reports before and after the reversal time when a reversal occurs, and to parse the key assumptions in the supporting data of the two conflict reports respectively. The event logging module is used to generate opinion reversal event logs based on changes in the key assumptions.

Images