A task text research and judgment method, device and equipment and storage medium

Abstract

The task text research and judgment method, device, equipment and storage medium provided by the embodiments of the present disclosure, the method comprises: decomposing a task text to be researched and judged into a sub-retrieval intention, generating a directional retrieval expression in combination with search engine features, and obtaining webpage data; identifying a key sentence and extracting a key objective entity index based on the webpage data through an information dehydration end; obtaining supporting data and performing three-dimensional verification based on the key sentence and the key objective entity index, outputting high-confidence data; triggering global knowledge base retrieval to obtain macro-environment data based on the high-confidence data; converting the macro-environment data into a weight factor to correct baseline data and high-confidence data synchronously, and obtaining a core gap by difference; and performing game deduction according to a correction result, the core gap and the macro-environment data to drive a deduction model, and finally generating a research and judgment report. The method realizes multi-language weak signal full-coverage capture and marketing language stripping, and forms an automatic closed loop from collection to decision-making.

Description

Technical Field

[0001] This disclosure relates to the field of information retrieval and processing technology, and in particular to a method, apparatus, device and storage medium for analyzing task text. Background Technology

[0002] Traditional Open Source Intelligence (OSINT) analysis and commercial / defense competitor research primarily rely on intelligence analysts manually conducting information retrieval, data cleaning, and analysis across the entire internet. With the explosive growth of internet data, traditional methods suffer from the following significant drawbacks: manual retrieval struggles to cover the vast amounts of multilingual vertical media, think tank reports, and academic journals globally, easily overlooking crucial weak signals. Existing automated information scraping systems typically only perform simple data aggregation based on keywords, failing to identify and filter corporate public relations articles and marketing rhetoric, and even less capable of finding independent sources for cross-validation of key indicators for the same event. Existing large-scale model information summarization tools often only summarize surface facts, lacking the ability to place them in a macro-environment for benchmarking analysis and extrapolation, and unable to directly output action recommendations with decision-making value. For specific entities (such as competitors), existing intelligence analysis is often one-off, unable to automatically build long-term historical baselines, resulting in unfounded extrapolation results. Summary of the Invention

[0003] To solve the above-mentioned technical problems, or at least partially solve them, this disclosure provides a method, apparatus, device, and storage medium for task text analysis.

[0004] This disclosure provides a method for analyzing task text, the method comprising: The text of the task to be analyzed is broken down into several sub-search intentions, and a targeted search expression is generated based on the sub-search intentions and the engine characteristics of the preset search engine. Web page data is obtained based on the targeted search expression. Based on the preset information dehydration end, key sentences in the web page data are identified, and key objective entity indicators are extracted from the key sentences. Supporting data is obtained based on the key objective entity indicators, and the supporting data is subjected to three-dimensional verification to obtain high confidence data. The high confidence data includes core entities and domain tags. Based on the high-confidence data, macro-environmental data is retrieved from a preset global knowledge base. The macro-environmental data is used as a weighting factor to correct the high-confidence data and the preset baseline data. The core gap is calculated based on the correction results. The correction results include the baseline correction results and the high-confidence correction results. Based on the preset simulation model, the task to be assessed is simulated according to the correction results, core gaps and macro-environment data, and an assessment report of the task to be assessed is obtained.

[0005] The method provided in this disclosure also includes: Store the assessment report in a preset archive; Based on the high confidence correction results in the assessment report, the core entity retrieves the historical baseline data of the core entity from the preset archive, and performs time-series evolution on the historical baseline data to obtain the historical high confidence correction results of the core entity. By comparing the historical high-confidence correction results with the high-confidence correction results, the logical mutation points of the core entity are identified; The logical mutation points are classified, and the weight factors are adjusted based on a preset category weight table.

[0006] The method provided in this disclosure decomposes the text of the task to be analyzed into several sub-search intentions, generates targeted search expressions based on the sub-search intentions and the engine features of a preset search engine, and obtains web page data based on the targeted search expressions, including: Extract entity words from the text of the task to be analyzed, and match key variables from a preset variable library based on the entity words; The key variables are combined to obtain several hypothetical propositions. Based on the hypothetical propositions, the text of the task to be evaluated is decomposed into several sub-retrieval intentions. The number of hypothetical propositions is the same as the number of sub-retrieval intentions. The sub-search intents are effectively filtered based on a preset source credibility classification database to obtain the filtering results; Based on the engine features of the preset search engine and the filtering results, a targeted retrieval expression is generated, and web page data is obtained based on the targeted retrieval expression.

[0007] The method provided in this disclosure identifies key statements in web page data based on a preset information dehydration process, extracts key objective entity indicators from the key statements, obtains supporting data based on the key objective entity indicators, and performs three-dimensional verification on the supporting data to obtain high-confidence data, including: The web page data is segmented into independent sentence sequences according to punctuation marks, and semantic vectors are generated based on the independent sentence sequences. Based on the preset information dehydration end, the semantic vector is semantically scored to obtain the semantic score corresponding to each semantic vector. The semantic vectors with semantic scores exceeding the preset scoring threshold are stripped to obtain key sentences. Based on the preset indicator table, key objective entity indicators are extracted from the key statements. Based on the engine features, the entities, actions, and values ​​in the key objective entity indicators are decomposed to obtain the verification expression. Based on the verification expression, supporting data is obtained, and supporting data from the same data source are merged according to the preset domain name attribution mapping table to obtain the merged result; The credibility of the data source corresponding to the merged result is evaluated based on a preset source credibility classification library. Data sources with credibility higher than the preset credibility and their corresponding merged results are retained to obtain the retained result. The retained results are validated in three dimensions based on entity, action, and numerical dimensions to obtain high-confidence data.

[0008] The method provided in this disclosure performs three-dimensional verification on the retained results based on entity dimension, action dimension, and numerical dimension to obtain high-confidence data, including: Based on a preset entity recognition model, extract the verification subject corresponding to each piece of supporting data in the retained results and the key objective entity indicators in the entity dimension; Extract the action predicates from the supporting data containing the verification subject, determine the semantic direction of the action predicates based on the key objective entity indicators in the action dimension corresponding to the action direction, and filter the supporting data in the retained results according to the semantic direction determination results to obtain the filtered supporting data. If the action predicate of the supporting data is in the same direction as the action dimension in the key objective entity indicator, then the action is deemed reasonable and the supporting data is retained. If the action predicate in the supporting data is opposite to or unrelated to the action dimension in the key objective entity indicator, it is determined to be an action conflict, and the supporting data is removed. Extract the numerical range of the filtered supporting data, determine the range of the numerical range and the corresponding indicator values ​​of the key objective entity indicators in the numerical dimension, obtain the range determination result, and further filter the filtered supporting data based on the range determination result to obtain three-dimensional compliance data. If the range determination result indicates a contradiction in magnitude, the filtered supporting data is removed; if there is no contradiction in magnitude, the filtered supporting data is retained. The number of independent information sources corresponding to the three-dimensional compliance data is counted, and the confidence level of the corresponding independent information source is calculated based on the number of independent information sources and preset rules. Based on the calculation results, high-confidence data in the three-dimensional compliance data is determined.

[0009] The method provided in this disclosure retrieves macro-environmental data from a preset global knowledge base based on the high-confidence data, uses the macro-environmental data as a weighting factor to correct the high-confidence data and preset baseline data, and calculates the core gap based on the correction result, including: Extract the core entities and domain tags from the high-confidence data, determine the macro-dimensional set corresponding to the domain tags based on a preset global knowledge base, obtain the structured historical data corresponding to the core entities and macro-dimensional set from the preset global knowledge base, generate an environment retrieval expression based on the core entities and macro-dimensional set, and obtain the current hard data based on the environment retrieval expression; The current hard data and structured historical data are combined to obtain macro-environmental data, and weighting factors are generated based on the macro-environmental data. The baseline data and the high-confidence data are corrected according to the weighting factors to obtain the baseline correction result and the high-confidence correction result. The core difference is obtained by subtracting the baseline correction result from the high-confidence correction result.

[0010] The method provided in this disclosure, based on a preset deduction model, deduces the task to be assessed according to the correction results, core gaps, and macro-environmental data, and obtains an assessment report of the task to be assessed, including: A macro-environment summary is determined based on the aforementioned macro-environment data; Based on the preset template, the correction results, core gaps and macro-environment summary are assembled to obtain inference prompts; The deduction prompt is input into a preset deduction model, the hypothetical purpose of the deduction prompt is output based on the preset deduction model, and an action path is generated based on the hypothetical purpose, wherein one hypothetical purpose corresponds to at least two action paths; The probability of occurrence of all action paths is calculated based on the preset inference model. The action paths are then sorted in descending order according to the probability of occurrence. The first action path and the hypothetical objective corresponding to the first action path are output to obtain the basis for countermeasures. Based on the preset deduction model, several countermeasure action suggestions are generated according to the countermeasure basis, and the countermeasure action suggestions are deduced to obtain the deduction results; An assessment report is generated by combining the text of the task to be assessed, the correction results, the core gaps, the summary of the macro environment, the basis for countermeasures, and the deduction results.

[0011] This disclosure also provides a task text analysis device, the device comprising: The acquisition module is used to decompose the text of the task to be analyzed into several sub-retrieval intentions, generate targeted retrieval expressions based on the sub-retrieval intentions and the engine features of the preset search engine, and acquire web page data based on the targeted retrieval expressions; The extraction module is used to identify key statements in the web page data based on preset information dehydration end, extract key objective entity indicators from the key statements, obtain supporting data based on the key objective entity indicators, and perform three-dimensional verification on the supporting data to obtain high confidence data. The calculation module is used to retrieve macro-environmental data from a preset global knowledge base based on the high-confidence data, use the macro-environmental data as a weighting factor to correct the high-confidence data and the preset baseline data, and calculate the core gap based on the correction result. The correction result includes the baseline correction result and the high-confidence correction result. The deduction module is used to deduce the task to be assessed based on the correction results, core gaps and macro-environment data, and obtain an assessment report of the task to be assessed.

[0012] The apparatus provided in this disclosure also includes: The feedback module is used to store the assessment report in a preset archive, retrieve the historical baseline data of the core entity based on the high confidence correction result in the assessment report from the preset archive, perform time-series evolution on the historical baseline data to obtain the historical high confidence correction result of the core entity, compare the historical high confidence correction result with the high confidence correction result, identify the logical mutation points of the core entity, classify the logical mutation points, and adjust the weight factors based on a preset category weight table.

[0013] The apparatus provided in this disclosure, wherein the acquisition module is specifically used for: Extract entity words from the text of the task to be analyzed, and match key variables from a preset variable library based on the entity words; The key variables are combined to obtain several hypothetical propositions. Based on the hypothetical propositions, the text of the task to be evaluated is decomposed into several sub-retrieval intentions. The number of hypothetical propositions is the same as the number of sub-retrieval intentions. The sub-search intents are effectively filtered based on a preset source credibility classification database to obtain the filtering results; Based on the engine features of the preset search engine and the filtering results, a targeted retrieval expression is generated, and web page data is obtained based on the targeted retrieval expression.

[0014] The extraction module in the apparatus provided in this disclosure is specifically used for: The web page data is segmented into independent sentence sequences according to punctuation marks, and semantic vectors are generated based on the independent sentence sequences. Based on the preset information dehydration end, the semantic vector is semantically scored to obtain the semantic score corresponding to each semantic vector. The semantic vectors with semantic scores exceeding the preset scoring threshold are stripped to obtain key sentences. Based on the preset indicator table, key objective entity indicators are extracted from the key statements. Based on the engine features, the entities, actions, and values ​​in the key objective entity indicators are decomposed to obtain the verification expression. Based on the verification expression, supporting data is obtained, and supporting data from the same data source are merged according to the preset domain name attribution mapping table to obtain the merged result; The credibility of the data source corresponding to the merged result is evaluated based on a preset source credibility classification library. Data sources with credibility higher than the preset credibility and their corresponding merged results are retained to obtain the retained result. The retained results are validated in three dimensions based on entity, action, and numerical dimensions to obtain high-confidence data.

[0015] The extraction module in the apparatus provided in this disclosure is specifically used for: Based on a preset entity recognition model, extract the verification subject corresponding to each piece of supporting data in the retained results and the key objective entity indicators in the entity dimension; Extract the action predicates from the supporting data containing the verification subject, determine the semantic direction of the action predicates based on the key objective entity indicators in the action dimension corresponding to the action direction, and filter the supporting data in the retained results according to the semantic direction determination results to obtain the filtered supporting data. If the action predicate of the supporting data is in the same direction as the action dimension in the key objective entity indicator, then the action is deemed reasonable and the supporting data is retained. If the action predicate in the supporting data is opposite to or unrelated to the action dimension in the key objective entity indicator, it is determined to be an action conflict, and the supporting data is removed. Extract the numerical range of the filtered supporting data, determine the range of the numerical range and the corresponding indicator values ​​of the key objective entity indicators in the numerical dimension, obtain the range determination result, and further filter the filtered supporting data based on the range determination result to obtain three-dimensional compliance data. If the range determination result indicates a contradiction in magnitude, the filtered supporting data is removed; if there is no contradiction in magnitude, the filtered supporting data is retained. The number of independent information sources corresponding to the three-dimensional compliance data is counted, and the confidence level of the corresponding independent information source is calculated based on the number of independent information sources and preset rules. Based on the calculation results, high-confidence data in the three-dimensional compliance data is determined.

[0016] The apparatus provided in this disclosure, wherein the computing module is specifically used for: Extract the core entities and domain tags from the high-confidence data, determine the macro-dimensional set corresponding to the domain tags based on a preset global knowledge base, obtain the structured historical data corresponding to the core entities and macro-dimensional set from the preset global knowledge base, generate an environment retrieval expression based on the core entities and macro-dimensional set, and obtain the current hard data based on the environment retrieval expression; The current hard data and structured historical data are combined to obtain macro-environmental data, and weighting factors are generated based on the macro-environmental data. The baseline data and the high-confidence data are corrected according to the weighting factors to obtain the baseline correction result and the high-confidence correction result. The core difference is obtained by subtracting the baseline correction result from the high-confidence correction result.

[0017] The apparatus provided in this disclosure, wherein the deduction module is specifically used for: A macro-environment summary is determined based on the aforementioned macro-environment data; Based on the preset template, the correction results, core gaps and macro-environment summary are assembled to obtain inference prompts; The deduction prompt is input into a preset deduction model, the hypothetical purpose of the deduction prompt is output based on the preset deduction model, and an action path is generated based on the hypothetical purpose, wherein one hypothetical purpose corresponds to at least two action paths; The probability of occurrence of all action paths is calculated based on the preset inference model. The action paths are then sorted in descending order according to the probability of occurrence. The first action path and the hypothetical objective corresponding to the first action path are output to obtain the basis for countermeasures. Based on the preset deduction model, several countermeasure action suggestions are generated according to the countermeasure basis, and the countermeasure action suggestions are deduced to obtain the deduction results; An assessment report is generated by combining the text of the task to be assessed, the correction results, the core gaps, the summary of the macro environment, the basis for countermeasures, and the deduction results.

[0018] This disclosure also provides an electronic device, the electronic device comprising: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the task text analysis method provided in this disclosure.

[0019] This disclosure also provides a computer-readable storage medium storing a computer program for executing the task text analysis method provided in this disclosure.

[0020] The technical solution provided in this disclosure has the following advantages compared with the prior art: This disclosure provides a task text analysis method that breaks down the task text to be analyzed into sub-search intentions, generates targeted search expressions based on the corresponding engine features of a search engine, obtains webpage data, identifies key statements in the webpage data through an information dehydration process, and extracts key objective entity indicators from these statements. Supporting data is obtained based on these key objective entity indicators, and three-dimensional verification is performed on the supporting data to obtain high-confidence data containing core entities and domain tags. A global knowledge base search is performed based on the high-confidence data to obtain macro-environment data, which is then converted into weight factors. Baseline data and high-confidence data are corrected according to these weight factors to obtain baseline correction results and high-confidence correction results. The difference is calculated to generate a core gap. Based on the correction results, the core gap, and the macro-environment data, an analysis report for the task to be analyzed is obtained. Through multilingual targeted retrieval and information dehydration, the blind spots of manual retrieval and the interference of advertorial noise are overcome. Three-dimensional verification and macro-weighted benchmarking compensate for the lack of cross-validation and in-depth analysis. A memory update mechanism solves the problem of inferences lacking a basis due to missing historical baselines. Ultimately, this achieves fully automated intelligence analysis from data collection to decision-making recommendations. The information dehydration and mandatory cross-verification mechanism for the intelligence field effectively eliminates the errors of single sources and the interference of public relations articles, significantly improving the objectivity and confidence of intelligence facts. Attached Figure Description

[0021] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0022] Figure 1 A flowchart illustrating a task text analysis method provided in this embodiment of the disclosure; Figure 2 This is a schematic diagram of the structure of a task text analysis device provided in an embodiment of the present disclosure; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0023] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0024] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0025] The terminology used herein, including but not limited to, is open-ended. The term "based on" is at least partially based on. The term "one embodiment" refers to at least one embodiment; the term "another embodiment" refers to at least one additional embodiment; the term "some embodiments" refers to at least some embodiments. Relevant definitions for other terms will be given in the description below.

[0026] It should be noted that the concepts of "first," "second," etc., mentioned in this disclosure are used only to distinguish different devices, modules, or units, and are not used to define the order of functions performed by these devices, modules, or units or their interdependencies.

[0027] It should be noted that the one or more modifications mentioned in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that they should be understood as one or more unless explicitly indicated in the context.

[0028] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0029] To address the aforementioned issues, this disclosure provides a method for analyzing task text, which will be described below with reference to specific embodiments.

[0030] Figure 1 This is a flowchart illustrating a task text analysis method provided in an embodiment of the present disclosure. The method can be executed by a task text analysis device, which can be implemented in software and / or hardware and is generally integrated into an electronic device.

[0031] Example 1: A task text analysis method provided in this embodiment includes: S101: Decompose the text of the task to be analyzed into several sub-retrieval intentions, and generate targeted retrieval expressions based on the sub-retrieval intentions and the engine features of the preset search engine, and obtain web page data based on the targeted retrieval expressions; S102: Based on the preset information dehydration end, identify key sentences in the web page data, extract key objective entity indicators from the key sentences, obtain supporting data based on the key objective entity indicators, and perform three-dimensional verification on the supporting data to obtain high confidence data, the high confidence data including core entities and domain tags; S103: Retrieve macro-environmental data from a preset global knowledge base based on the high-confidence data, use the macro-environmental data as a weighting factor to correct the high-confidence data and the preset baseline data, calculate the core gap based on the correction result, and the correction result includes the baseline correction result and the high-confidence correction result; S104: Based on the preset simulation model, the task to be assessed is simulated according to the correction results, core gaps and macro-environment data to obtain the assessment report of the task to be assessed.

[0032] In this embodiment, the task text to be analyzed is the original intelligence request instruction received by the system. Through semantic deconstruction and entity extraction, it is broken down into several sub-retrieval intentions. Each sub-retrieval intention is a verifiable factual direction after task decomposition, which is combined with the engine features of a preset search engine to generate a targeted retrieval expression. Engine features are the Boolean logic, language restrictions, and other grammatical rules supported by the search engine, which are combined and mapped with the filtered sub-retrieval intentions to generate the targeted retrieval expression. The targeted retrieval expression is a machine retrieval instruction with source direction, which is submitted to the search engine and returns web page data. The web page data is multilingual unstructured raw material, input into a preset information dehydration module for noise reduction and semantic scoring. The preset information dehydration module is a semantic processing module that removes marketing rhetoric, identifying key statements from the web page data. Key statements are objective factual descriptions after removing subjective modifications, from which key objective entity indicators are extracted through named entity recognition and relation extraction. Key objective entity indicators are structured factual fields containing entities, actions, and values, which are used to generate verification expressions and retrieve supporting data. Supporting data consists of verification evidence from independent sources, aggregated into high-confidence data after being filtered through three-dimensional verification of entities, actions, and values. The three-dimensional verification sequentially determines whether the supporting data aligns with the original indicators in terms of the subject of the argument, the direction of the action, and the scale. High-confidence data is a collection of credible facts verified by multiple independent sources, containing core entities and domain labels. Core entities are the identified analytical objects, and domain labels are the industry classifications to which the core entities belong. Together, they trigger macro-dimensional matching and retrieval in a pre-set global knowledge base. The pre-set global knowledge base is a knowledge engine that stores structured historical data and dimensional mapping relationships, returning macro-environmental data. Macro-environmental data includes external environmental information such as hard conditions, budgets, and regulations, which is quantified to generate weighting factors. Weighting factors are calculable coefficients of macro-level impact, simultaneously applied to the pre-set baseline data and high-confidence data to perform corrections. Baseline data represents our static capability benchmark value; after correction by weighting factors, the baseline correction result is obtained. The high-confidence correction result reflects the actual level of the opponent's current capabilities under the macro-environment. The difference between the corrected results yields the core gap, which is the effective capability difference after macro-weighting. The corrected results, the core gap, and macro-environmental data are combined to form inference prompts, which drive the preset inference model to perform multi-branch game simulation and strategy evaluation, ultimately generating an analysis report containing situation assessments and action recommendations.

[0033] Specifically, the task text received by the system is to analyze the latest developments of Company A (a competitor) in the field of seeker heads. Through semantic deconstruction and variable combination, the task is broken down into multiple sub-search intentions, such as production capacity changes, technical parameters, and supply chain adjustments. Combining the engine characteristics of the preset search engine (such as Boolean logic and site-limited syntax), the sub-search intentions are transformed into targeted search expressions such as site:defense.gov Company A AND seeker head, which are submitted to the search engine and return multiple news reports, i.e., web page data. The web page data contains original sentences with subjective modifications, such as "Company A's seeker head performance is the best in the world." After being processed by a preset information dehydration process, marketing rhetoric is removed and semantic scoring is performed. Noise exceeding the threshold of "the best in the world" is eliminated, retaining the objective fact that Company A plans to increase the production capacity of its X-type seeker head by 4 times, i.e., the key statement. From the key statement, named entity recognition is used to extract {entity: Company A X-type seeker head, action: increase, value: 4 times}, i.e., the key objective entity indicator. Based on this indicator, the system automatically generates a verification expression and initiates a secondary search to obtain supporting data from financial reports of Journal B and Media C. The supporting data, after undergoing three-dimensional verification—entity matching (confirming the subject of the argument is Company A's X-type seeker head), action direction determination (confirming both are capacity expansions rather than reductions), and magnitude comparison (confirming consistent capacity multiples)—was deemed to have passed verification and was aggregated into high-confidence data. This high-confidence data includes the core entity Company A and the domain-labeled seeker head / preset guidance system. These two elements jointly trigger macro-dimensional matching in the preset global knowledge base, retrieving macro-environmental data such as budget growth prevention and material control. This macro-environmental data is quantified and transformed into weighting factors, simultaneously affecting the preset baseline and high-confidence data. After weighted correction, the baseline correction result and the high-confidence correction result are obtained, and the difference between the two generates the core gap. The correction result, the core gap, and the macro-environmental data are assembled into deduction prompts, driving the preset deduction model to perform game simulation and strategy evaluation. The deduction concludes that the 4-fold capacity expansion is not only a commercial activity but also a strategic preparedness action. Finally, an analysis report is generated, including action recommendations such as suggesting that our side advance supply chain backup, and Company A's expanded production of X-type seeker head is written into a memory file for future retrieval.

[0034] The working principle and beneficial effects of this embodiment are as follows: After decomposing the text of the task to be analyzed into sub-search intentions, a targeted search expression is generated by combining the engine features of the corresponding search engine to obtain webpage data. Key statements in the webpage data are determined through an information dehydration process, and key objective entity indicators are extracted from these statements. Supporting data is obtained based on these key objective entity indicators. Three-dimensional verification is performed on the supporting data to obtain high-confidence data containing core entities and domain tags. A global knowledge base search is performed based on the high-confidence data to obtain macro-environmental data, which is then converted into weighting factors. Baseline data and high-confidence data are corrected according to these weighting factors to obtain baseline correction results and high-confidence correction results. The difference is calculated to generate a core gap. Based on the correction results, the core gap, and the macro-environmental data, an analysis report for the task to be analyzed is obtained. Through multilingual targeted retrieval and information dehydration, the blind spots of manual retrieval and the interference of soft-sell noise are overcome. Three-dimensional verification and macro-weighted benchmarking compensate for the lack of cross-validation and in-depth analysis. The memory update mechanism solves the problem of unfounded analysis due to missing historical baselines. Ultimately, a fully automated intelligence analysis process from data collection to decision-making recommendations is achieved. The information dehydration and mandatory cross-verification mechanism for the intelligence field effectively eliminates the errors of single sources and the interference of public relations articles, significantly improving the objectivity and confidence of intelligence facts.

[0035] Example 2: The method provided in this embodiment of the present disclosure further includes: Store the assessment report in a preset archive; Based on the high confidence correction results in the assessment report, the core entity retrieves the historical baseline data of the core entity from the preset archive, and performs time-series evolution on the historical baseline data to obtain the historical high confidence correction results of the core entity. By comparing the historical high-confidence correction results with the high-confidence correction results, the logical mutation points of the core entity are identified; The logical mutation points are classified, and the weight factors are adjusted based on a preset category weight table.

[0036] In this embodiment, the preset archive is a persistent memory repository storing past assessment reports. Each record is archived by core entity and includes a timestamp and confidence level tag. After the current assessment report is stored, the system retrieves the historical baseline data of the entity from the preset archive using the core entity in the high-confidence correction result as the search key. The historical baseline data is a sequence of high-confidence facts confirmed by cross-validation in past assessments of the entity, including snapshots of past capacity changes, actions, and corresponding macroeconomic environments. The historical baseline data is subjected to time-series evolution, that is, the trend and magnitude of indicator changes in adjacent assessment periods are extracted, recurring action-result correlation patterns are identified, and historical high-confidence correction results are summarized and output. The historical high-confidence correction results are a numerical expression of the entity's historical behavioral patterns. For example, in the past three years, the actual production capacity reached after each announced expansion was delayed by 6 months. The historical high-confidence correction results are compared with the current high-confidence correction results in terms of both value and direction, and the system automatically identifies logical abrupt change points. Logical mutation points are signals where an entity's current behavior deviates significantly from historical patterns. For example, in a recent capacity expansion, the actual capacity fulfillment might exceed expectations. Logical mutation points are categorized according to a pre-defined classification system, including categories such as sudden changes in capacity fulfillment rate, supply chain strategy shifts, and technology path transitions. The classification results trigger a query of a pre-defined category weight table, which stores the mapping relationship and adjustment coefficients between mutation categories and inference model parameters. Based on the query results, feedback adjustments are performed on the weight factors in the inference model, increasing or decreasing the default values ​​of corresponding parameters to achieve dynamic self-correction of the inference logic.

[0037] Specifically, after this assessment, the system stores the assessment report, which includes conclusions such as Company A's X-type seeker head production having increased fourfold, the core gap being that our production capacity is only 25% of Company A's, and recommendations to promote supply chain backup, in a pre-set archive. Subsequently, the system initiates a query based on Company A, the core entity in the high-confidence correction results, retrieving historical baseline data for this entity from the pre-set archive. This includes historical facts recorded in the past three assessments, such as Company A's actual production reaching its target lag by 6 to 8 months after each announced expansion, and the typically 30% reduction in advertised production capacity. These historical baseline data are then subjected to time-series evolution analysis to extract the trend of its production capacity fulfillment rate. The analysis concludes that the historical average production capacity fulfillment rate for Company A under the high-confidence correction results is 70%, and there has never been an overfulfillment. Comparing this historical pattern with the current high-confidence correction result of a fourfold increase in production capacity, which has been verified by multiple sources to have actually reached its target, the system identifies a logical turning point: the production capacity fulfillment rate has jumped from 70% to 100%, indicating a fundamental shift in the behavioral pattern. The mutation point was automatically classified as a positive mutation in capacity fulfillment rate, triggering a query of a preset category weight table. This table stores the mapping relationship between the capacity fulfillment rate mutation and the competitor's capacity loss coefficient in the inference model. Based on the adjustment coefficient returned by the query, the system performs feedback adjustment on the competitor's capacity loss coefficient in the weight factors, reducing it from the historical default of 0.7 to 0.0, so that the subsequent inference model's prediction of this entity is more in line with its current behavior pattern.

[0038] The working principle and beneficial effects of this embodiment are as follows: After storing the assessment report in a preset archive, historical baseline data is retrieved based on the core entities in the high-confidence correction results. The historical baseline data is then used to generate historical high-confidence correction results through time-series evolution. The historical high-confidence correction results are compared with the current high-confidence correction results to identify logical mutation points. After classification, the logical mutation points trigger a query of a preset category weight table, and feedback adjustments are made to the weight factors to complete the dynamic self-correction of the inference logic. By storing previous assessment reports in a preset archive and retrieving historical baseline data for time-series evolution, and comparing and identifying logical mutation points to make feedback adjustments to the weight factors, the deficiency of single analysis being unable to accumulate historical knowledge is overcome. This allows the inference model to adaptively evolve with changes in entity behavior patterns, improving the accuracy and predictive reliability of subsequent assessments.

[0039] Example 3: The method provided in this embodiment of the present disclosure decomposes the text of the task to be analyzed into several sub-search intentions, generates targeted search expressions based on the sub-search intentions and the engine features of a preset search engine, and obtains web page data based on the targeted search expressions, including: Extract entity words from the text of the task to be analyzed, and match key variables from a preset variable library based on the entity words; The key variables are combined to obtain several hypothetical propositions. Based on the hypothetical propositions, the text of the task to be evaluated is decomposed into several sub-retrieval intentions. The number of hypothetical propositions is the same as the number of sub-retrieval intentions. The sub-search intents are effectively filtered based on a preset source credibility classification database to obtain the filtering results; Based on the engine features of the preset search engine and the filtering results, a targeted retrieval expression is generated, and web page data is obtained based on the targeted retrieval expression.

[0040] In this embodiment, entity words are specific object names extracted from the text of the task to be analyzed through named entity recognition, such as company names, product models, or technical terms, which constitute the search keys for subsequent variable matching. After entity words are input into a preset variable library, key variables are matched and output. The preset variable library is a structured dictionary that stores the mapping relationship between entities and competitive analysis dimensions. Key variables are quantifiable and traceable analysis dimension labels such as production capacity changes, technical parameters, and supply chain adjustments. Key variables are combined through competitive logic to generate hypothetical propositions. Hypothetical propositions are verifiable conjectures formed by adversarially associating isolated variables, such as Company A's capacity expansion behavior being a strategic preparation rather than a commercial order. The number of hypothetical propositions determines the granularity of the sub-search intent. Each hypothetical proposition is directly mapped to a sub-search intent, ensuring that the decomposition results correspond one-to-one with the verification target. Sub-search intents are input into a preset source credibility rating library for filtering. The preset source credibility rating library is an evaluation matrix that stores the authority ratings of various types of sources. The filtering operation removes sub-search intents that cannot be mapped to any high-level source, and outputs the filtering results. The filtered results are combined and mapped with the engine features of the preset search engine. The engine features are the Boolean operators, language-specific syntax, and site filtering rules supported by the search engine. The combination mapping transforms the entities and actions in the filtered results into legal query strings that conform to the engine syntax, generates a targeted search expression, and returns web page data after the targeted search expression is delivered.

[0041] The working principle and beneficial effects of this embodiment are as follows: Entity words are extracted from the text of the task to be analyzed, and key variables are matched from a preset variable library based on these entity words. Key variable combinations generate hypothetical propositions, which drive the decomposition of the text of the task to be analyzed into sub-retrieval intentions, with the number of hypothetical propositions and sub-retrieval intentions being consistent. Sub-retrieval intentions are filtered by a preset source credibility grading library, eliminating intentions that cannot map to high-level sources, resulting in a filtering result. The filtering result is mapped with preset search engine feature combinations to generate a targeted retrieval expression, and webpage data is obtained based on this targeted retrieval expression. By extracting entity words and combining variables to generate hypothetical propositions, fuzzy tasks are decomposed into verifiable sub-retrieval intentions. After filtering by the source credibility grading library and mapping with search engine features to generate targeted retrieval expressions, the shortcomings of incomplete coverage and blind retrieval direction in manual retrieval are overcome, improving the efficiency of targeted capture of high-value multilingual sources and the accuracy of intelligence gathering.

[0042] Example 4: The method provided in this embodiment of the present disclosure identifies key statements in the webpage data based on a preset information dehydration process, extracts key objective entity indicators from the key statements, obtains supporting data based on the key objective entity indicators, and performs three-dimensional verification on the supporting data to obtain high-confidence data, including: The web page data is segmented into independent sentence sequences according to punctuation marks, and semantic vectors are generated based on the independent sentence sequences. Based on the preset information dehydration end, the semantic vector is semantically scored to obtain the semantic score corresponding to each semantic vector. The semantic vectors with semantic scores exceeding the preset scoring threshold are stripped to obtain key sentences. Based on the preset indicator table, key objective entity indicators are extracted from the key statements. Based on the engine features, the entities, actions, and values ​​in the key objective entity indicators are decomposed to obtain the verification expression. Based on the verification expression, supporting data is obtained, and supporting data from the same data source are merged according to the preset domain name attribution mapping table to obtain the merged result; The credibility of the data source corresponding to the merged result is evaluated based on a preset source credibility classification library. Data sources with credibility higher than the preset credibility and their corresponding merged results are retained to obtain the retained result. The retained results are validated in three dimensions based on entity, action, and numerical dimensions to obtain high-confidence data.

[0043] In this embodiment, the independent sentence sequence is the smallest semantically complete text unit obtained by segmenting web page data according to punctuation marks, retaining the pure text content after eliminating HTML tags and navigation bar noise. The independent sentence sequence is input into a pre-trained language model to generate semantic vectors. The semantic vector is a distributed numerical representation of the semantic context of the sentence, carrying the deep logical meaning of the sentence and providing a computational foundation for subsequent marketing rhetoric recognition. The preset scoring threshold is the discrimination boundary value that distinguishes objective facts from marketing rhetoric. It is determined through model optimization on the labeled training set. After the semantic vector is processed by the preset information dehydration end to calculate the semantic score, sentences with scores exceeding the preset scoring threshold are judged as high-concentration marketing rhetoric and are stripped, while sentences that do not exceed the threshold are retained as key statements. The preset indicator table is a structured pattern library that stores domain-specific indicator templates, defining the types and unit specifications of extractable indicators such as capacity, accuracy, and delivery cycle. Key statements are matched with the preset indicator table to extract key objective entity indicators. In this stage, the engine features reuse the Boolean syntax and field limitation rules of the search engine to decompose the key objective entity indicators into multiple verification expressions from different angles and information sources. The verification expressions are machine instructions that drive secondary retrieval to obtain supporting data. A pre-defined domain attribution mapping table maintains the attribution relationship between root domains and parent groups. Supporting data from different subdomains within the same media group are merged and deduplicated, outputting the merged result. A pre-set confidence level is the minimum admission score set in the source confidence grading database. Data sources with confidence levels below this score are discarded, while those above are retained. The retained results undergo layer-by-layer verification across entity, action, and numerical dimensions, filtering out data with inconsistent subject matter, conflicting actions, and contradictory magnitudes. Finally, the number of independent sources is aggregated, and the confidence level is calculated, outputting high-confidence data.

[0044] The working principle and beneficial effects of this embodiment are as follows: web page data is segmented into independent sentence sequences according to punctuation marks. Semantic vectors are generated from the independent sentence sequences. Semantic vectors are scored based on a preset information dehydration process. Semantic vectors with semantic scores exceeding a preset scoring threshold are stripped to obtain key sentences. Key objective entity indicators are extracted from the key sentences according to a preset indicator table. Entities, actions, and values ​​in the key objective entity indicators are decomposed based on engine features to obtain verification expressions. Supporting data is obtained based on the verification expressions. Supporting data from the same data source are merged using a preset domain name attribution mapping table to obtain a merged result. The system uses a pre-defined credibility grading database to assess the credibility of the merged data sources. It retains data sources with credibility higher than the pre-set credibility and their corresponding merged results, thus obtaining retained results. Based on entity, action, and numerical dimensions, the retained results undergo three-dimensional verification to obtain high-confidence data. Key statements are obtained by stripping marketing rhetoric through semantic scoring. Key objective entity indicators are extracted from the key statements and decomposed to generate verification expressions. After merging multi-source corroborating data and filtering the credibility of the sources, three-dimensional verification is performed. This overcomes the defects of noise interference from advertorials and the unreliability of a single source, realizes the automatic extraction of objective facts and independent verification from multiple sources, and produces intelligence primitives with high credibility.

[0045] Example 5: The method provided in this embodiment of the present disclosure performs three-dimensional verification on the retained results based on entity dimension, action dimension, and numerical dimension to obtain high-confidence data, including: Based on a preset entity recognition model, extract the verification subject corresponding to each piece of supporting data in the retained results and the key objective entity indicators in the entity dimension; Extract the action predicates from the supporting data containing the verification subject, determine the semantic direction of the action predicates based on the key objective entity indicators in the action dimension corresponding to the action direction, and filter the supporting data in the retained results according to the semantic direction determination results to obtain the filtered supporting data. If the action predicate of the supporting data is in the same direction as the action dimension in the key objective entity indicator, then the action is deemed reasonable and the supporting data is retained. If the action predicate in the supporting data is opposite to or unrelated to the action dimension in the key objective entity indicator, it is determined to be an action conflict, and the supporting data is removed. Extract the numerical range of the filtered supporting data, determine the range of the numerical range and the corresponding indicator values ​​of the key objective entity indicators in the numerical dimension, obtain the range determination result, and further filter the filtered supporting data based on the range determination result to obtain three-dimensional compliance data. If the range determination result indicates a contradiction in magnitude, the filtered supporting data is removed; if there is no contradiction in magnitude, the filtered supporting data is retained. The number of independent information sources corresponding to the three-dimensional compliance data is counted, and the confidence level of the corresponding independent information source is calculated based on the number of independent information sources and preset rules. Based on the calculation results, high-confidence data in the three-dimensional compliance data is determined.

[0046] In this embodiment, the preset entity recognition model is a named entity recognizer trained with annotations. It extracts verification subjects from the supporting data that match the entity dimension of the key objective entity indicators. The verification subject is the core object of the supporting data discussion, and its correspondence with the entity dimension ensures that the source discussion and the original indicator are aligned. Action predicates are extracted from the supporting data containing verification subjects. Action predicates are verbs or verb phrases describing the actions performed by the entity. Semantic direction determination is performed between these predicates and the action direction corresponding to the key objective entity indicator in the action dimension. Action direction is the vector attribute of the action in the indicator, such as increase, decrease, or maintain. If the directions are the same, the action is considered reasonable and retained; if the directions are opposite or irrelevant, the action is considered conflicting and eliminated, outputting the filtered supporting data. Numerical ranges are extracted from the filtered supporting data. Numerical ranges are the specific numerical ranges disclosed in the supporting data. These ranges are compared with the indicator values ​​corresponding to the key objective entity indicators in the numerical dimension for range determination. Range determination checks for magnitude contradictions between the numerical ranges and indicator values ​​through magnitude comparison. Magnitude contradictions are irreconcilable conflicts between the two in terms of magnitude; if such contradictions exist, they are eliminated; otherwise, they are retained, outputting three-dimensional compliant data. The statistical count of independent sources corresponding to the three-dimensional compliance data is based on the count of independent parent groups after domain name attribution deduplication. The number of independent sources and preset rules are used together to calculate the confidence level. The preset rules are a scoring logic that combines a threshold for the number of independent sources with an authority-weighted average. The confidence level is a graded label for the credibility of the verification conclusion. Finally, those data with a confidence level that reaches the preset high level are selected from the three-dimensional compliance data and identified as high-confidence data.

[0047] The working principle and beneficial effects of this embodiment are as follows: A preset entity recognition model is used to extract the verification subject corresponding to each piece of supporting data and key objective entity indicators in the entity dimension from the retained results. Then, action predicates are extracted from the supporting data containing the verification subject. Based on the action direction in the action dimension, the semantic direction of the action predicates is determined. If the directions are the same, the action is deemed reasonable and retained; if the directions are opposite or irrelevant, the action is deemed conflicting and removed, thus obtaining filtered supporting data. Subsequently, the numerical range of the filtered supporting data is extracted and compared with the indicator values ​​in the numerical dimension for range determination. If there is a magnitude contradiction, the data is removed; if there is no magnitude contradiction, it is retained, thus obtaining three-dimensional compliant data. The number of independent information sources corresponding to the three-dimensional compliant data is counted, and the confidence level is calculated based on the number of independent information sources and preset rules, thereby determining high-confidence data. By matching entity dimensions to eliminate supporting data that does not match the subject of the argument, filtering out interfering sources of conflicting actions by determining the direction of actions, and eliminating erroneous data with contradictory magnitudes by determining the range of numerical values, the confidence level is finally calculated based on the number of independent sources. This overcomes the shortcomings of unreliability of a single source and inconsistency in data caliber, and improves the accuracy of intelligence verification and the credibility of conclusions.

[0048] Example 6: The method provided in this embodiment of the present disclosure retrieves macro-environmental data from a preset global knowledge base based on the high-confidence data, uses the macro-environmental data as a weighting factor to correct the high-confidence data and preset baseline data, and calculates the core gap based on the correction result, including: Extract the core entities and domain tags from the high-confidence data, determine the macro-dimensional set corresponding to the domain tags based on a preset global knowledge base, obtain the structured historical data corresponding to the core entities and macro-dimensional set from the preset global knowledge base, generate an environment retrieval expression based on the core entities and macro-dimensional set, and obtain the current hard data based on the environment retrieval expression; The current hard data and structured historical data are combined to obtain macro-environmental data, and weighting factors are generated based on the macro-environmental data. The baseline data and the high-confidence data are corrected according to the weighting factors to obtain the baseline correction result and the high-confidence correction result. The core difference is obtained by subtracting the baseline correction result from the high-confidence correction result.

[0049] In this embodiment, the domain label is a semantic annotation of the industry classification of the core entity in the high-confidence data. It is automatically classified from the industry attributes and product characteristics of the core entity and includes preset categories such as guided vehicles, new energy, and semiconductors. After the domain label is input into the preset global knowledge base, it is matched and output as a set of macro dimensions. The set of macro dimensions is a group of macro analysis perspectives bound to the domain label in the preset global knowledge base. The preset domain activates dimensions such as budget, export control, and geopolitical hard conditions. The new energy domain activates dimensions such as subsidy hard conditions, carbon emission hard conditions, and material price index. The structured historical data is the text of hard conditions, budget values, and summary of control lists stored in the preset global knowledge base by double-key indexing of core entities and macro dimensions. The structured historical data, core entities, and macro dimension set are used together to generate an environment retrieval expression. The environment retrieval expression is a macro data retrieval instruction with additional time constraints and information source orientation. After submission, the current hard data is obtained. The current hard data is the value and summary of the latest released hard condition bills, budget reports, or trade control documents. The current hard data and the set of structured historical data are then quantitatively decomposed by a large language model to generate weight factors. Weighting factors are applied simultaneously to both the preset baseline data and high-confidence data for correction. The baseline data, after weighting factor correction, yields the baseline-corrected result, and the high-confidence data, after weighting factor correction, yields the high-confidence-corrected result. The baseline-corrected result reflects our actual mobilization capacity under the macroeconomic environment, while the high-confidence-corrected result reflects the competitor's true achievement level under the macroeconomic environment. The difference between the two yields the core gap. The core gap is the effective capability difference after macroeconomic weighting, including attribution labels indicating whether the gap stems from production capacity scale, supply chain resilience, or favorable hard conditions.

[0050] The working principle and beneficial effects of this embodiment are as follows: Core entities and domain labels are extracted from high-confidence data. Then, a macro-dimensional set corresponding to the domain labels is determined based on a preset global knowledge base. Structured historical data corresponding to the core entities and macro-dimensional sets is obtained from the preset global knowledge base. Subsequently, an environment retrieval expression is generated based on the core entities and macro-dimensional sets, and current hard data is obtained based on the environment retrieval expression. The current hard data and the structured historical data set are combined to obtain macro-environment data. Weighting factors are generated based on the macro-environment data. Based on the weighting factors, preset baseline data and high-confidence data are corrected, resulting in baseline correction results and high-confidence correction results. The difference between the baseline correction results and the high-confidence correction results is calculated to obtain the core gap. By activating the macro-dimensional set through domain labels and obtaining historical and current macro data through dual channels, the macro environment is quantified as a weighting factor that simultaneously corrects the baseline data and high-confidence data before calculating the difference. This overcomes the shortcomings of static comparisons that ignore the influence of the external environment and the lack of contextual calibration of the data from both sides, ensuring that the core gap truly reflects the effective capability difference under the influence of macro variables.

[0051] Example 7: The method provided in this embodiment of the present disclosure, based on a preset inference model, infers the task to be assessed according to the correction results, core gaps, and macro-environmental data, to obtain an assessment report of the task to be assessed, including: A macro-environment summary is determined based on the aforementioned macro-environment data; Based on the preset template, the correction results, core gaps and macro-environment summary are assembled to obtain inference prompts; The deduction prompt is input into a preset deduction model, the hypothetical purpose of the deduction prompt is output based on the preset deduction model, and an action path is generated based on the hypothetical purpose, wherein one hypothetical purpose corresponds to at least two action paths; The probability of occurrence of all action paths is calculated based on the preset inference model. The action paths are then sorted in descending order according to the probability of occurrence. The first action path and the hypothetical objective corresponding to the first action path are output to obtain the basis for countermeasures. Based on the preset deduction model, several countermeasure action suggestions are generated according to the countermeasure basis, and the countermeasure action suggestions are deduced to obtain the deduction results; An assessment report is generated by combining the text of the task to be assessed, the correction results, the core gaps, the summary of the macro environment, the basis for countermeasures, and the deduction results.

[0052] In this embodiment, the macro-environment summary is a structured scenario description refined from macro-environment data using a large language model, focusing on core conclusions of key external variables such as budget changes and hard condition controls. The macro-environment summary, along with the revised results and core gaps, are assembled using a preset template to generate deduction prompts. The preset template is a structured framework that specifies the format of intelligence input and thought chain instructions. After the deduction prompts are input into the preset deduction model, the output is a hypothetical objective. This hypothetical objective represents the underlying motivation of the adversary to execute high-confidence facts, such as securing orders or squeezing market share. Each hypothetical objective drives the preset deduction model to generate action paths. These action paths are the specific sequence of actions the adversary might take in a future time window to achieve the hypothetical objective; each hypothetical objective corresponds to at least two action paths for game evaluation. The preset deduction model calculates the probability of all action paths occurring, estimates them based on historical baselines and macro trends, and then sorts them in descending order, outputting the first action path and its corresponding hypothetical objective as the basis for countermeasures. The basis for countermeasures is the combination of the highest threat path confirmed by the deduction and the adversary's motivation. Based on this, the preset deduction model matches and generates countermeasure action suggestions from the response strategy library. The proposed countermeasures are tested and verified for feasibility and consequences using a pre-set simulation model, and the simulation results are then output. The simulation results, along with the text of the task to be analyzed, the revised results, the core gaps, a summary of the macro environment, and a set of countermeasure justifications, are packaged together to generate an analysis report.

[0053] The working principle and beneficial effects of this embodiment are as follows: A macro-environment summary is determined based on macro-environment data. Then, based on a preset template, the correction results, core gaps, and the macro-environment summary are assembled to obtain deduction prompts. These prompts are input into a preset deduction model, which outputs the hypothetical purpose of the prompts. Action paths are generated based on these hypothetical purposes, with each hypothetical purpose corresponding to at least two action paths. The preset deduction model then calculates the probability of occurrence for all action paths, sorts them in descending order based on the probability, and outputs the first action path and its corresponding hypothetical purpose, thus obtaining the basis for countermeasures. Based on the preset deduction model and the basis for countermeasures, several countermeasure action suggestions are generated, and these suggestions are deduced to obtain the deduction results. Finally, the text of the task to be assessed, the correction results, the core gaps, the macro-environment summary, the basis for countermeasures, and the deduction results are combined to generate an assessment report. By summarizing and correcting the macro-environment and assembling the core gaps into deduction prompts, the deduction model is driven to output hypothetical objectives and generate multiple action paths. The first path is locked as the basis for countermeasures through probability ranking and countermeasure action suggestions are generated. This overcomes the shortcomings of traditional summary tools that only summarize facts and lack the ability to make deductions and provide decision support. It realizes the fully automated closed-loop output from intelligence gathering to action suggestions.

[0054] To implement the above embodiments, this disclosure also proposes a task text analysis device.

[0055] Figure 2 This is a schematic diagram of a task text analysis device provided in an embodiment of this disclosure. The device 200 can be implemented by software and / or hardware, and is generally integrated into an electronic device. Figure 2 As shown, the device 200 includes: an acquisition module 201, an extraction module 202, a calculation module 203, a deduction module 204, and a feedback module 205, wherein, The acquisition module 201 is used to decompose the text of the task to be analyzed into several sub-retrieval intentions, generate a targeted retrieval expression based on the sub-retrieval intentions and the engine features of the preset search engine, and acquire web page data based on the targeted retrieval expression; The extraction module 202 is used to identify key sentences in the web page data based on preset information dehydration end, extract key objective entity indicators from the key sentences, obtain supporting data based on the key objective entity indicators, and perform three-dimensional verification on the supporting data to obtain high confidence data. The calculation module 203 is used to retrieve macro-environmental data from a preset global knowledge base based on the high-confidence data, use the macro-environmental data as a weighting factor to correct the high-confidence data and the preset baseline data, and calculate the core gap based on the correction result. The correction result includes the baseline correction result and the high-confidence correction result. The deduction module 204 is used to deduce the task to be assessed based on the correction results, core gaps and macro-environment data, and obtain an assessment report of the task to be assessed.

[0056] The apparatus provided in this disclosure also includes: Feedback module 205 is used to store the assessment report in a preset archive, retrieve the historical baseline data of the core entity based on the high confidence correction result in the assessment report from the preset archive, perform time-series evolution on the historical baseline data to obtain the historical high confidence correction result of the core entity, compare the historical high confidence correction result with the high confidence correction result, identify the logical mutation points of the core entity, classify the logical mutation points, and adjust the weight factors based on a preset category weight table.

[0057] The apparatus provided in this disclosure, wherein the acquisition module is specifically used for: Extract entity words from the text of the task to be analyzed, and match key variables from a preset variable library based on the entity words; The key variables are combined to obtain several hypothetical propositions. Based on the hypothetical propositions, the text of the task to be evaluated is decomposed into several sub-retrieval intentions. The number of hypothetical propositions is the same as the number of sub-retrieval intentions. The sub-search intents are effectively filtered based on a preset source credibility classification database to obtain the filtering results; Based on the engine features of the preset search engine and the filtering results, a targeted retrieval expression is generated, and web page data is obtained based on the targeted retrieval expression.

[0058] The extraction module in the apparatus provided in this disclosure is specifically used for: The web page data is segmented into independent sentence sequences according to punctuation marks, and semantic vectors are generated based on the independent sentence sequences. Based on the preset information dehydration end, the semantic vector is semantically scored to obtain the semantic score corresponding to each semantic vector. The semantic vectors with semantic scores exceeding the preset scoring threshold are stripped to obtain key sentences. Based on the preset indicator table, key objective entity indicators are extracted from the key statements. Based on the engine features, the entities, actions, and values ​​in the key objective entity indicators are decomposed to obtain the verification expression. Based on the verification expression, supporting data is obtained, and supporting data from the same data source are merged according to the preset domain name attribution mapping table to obtain the merged result; The credibility of the data source corresponding to the merged result is evaluated based on a preset source credibility classification library. Data sources with credibility higher than the preset credibility and their corresponding merged results are retained to obtain the retained result. The retained results are validated in three dimensions based on entity, action, and numerical dimensions to obtain high-confidence data.

[0059] The extraction module in the apparatus provided in this disclosure is specifically used for: Based on a preset entity recognition model, extract the verification subject corresponding to each piece of supporting data in the retained results and the key objective entity indicators in the entity dimension; Extract the action predicates from the supporting data containing the verification subject, determine the semantic direction of the action predicates based on the key objective entity indicators in the action dimension corresponding to the action direction, and filter the supporting data in the retained results according to the semantic direction determination results to obtain the filtered supporting data. If the action predicate of the supporting data is in the same direction as the action dimension in the key objective entity indicator, then the action is deemed reasonable and the supporting data is retained. If the action predicate in the supporting data is opposite to or unrelated to the action dimension in the key objective entity indicator, it is determined to be an action conflict, and the supporting data is removed. Extract the numerical range of the filtered supporting data, determine the range of the numerical range and the corresponding indicator values ​​of the key objective entity indicators in the numerical dimension, obtain the range determination result, and further filter the filtered supporting data based on the range determination result to obtain three-dimensional compliance data. If the range determination result indicates a contradiction in magnitude, the filtered supporting data is removed; if there is no contradiction in magnitude, the filtered supporting data is retained. The number of independent information sources corresponding to the three-dimensional compliance data is counted, and the confidence level of the corresponding independent information source is calculated based on the number of independent information sources and preset rules. Based on the calculation results, high-confidence data in the three-dimensional compliance data is determined.

[0060] The apparatus provided in this disclosure, wherein the computing module is specifically used for: Extract the core entities and domain tags from the high-confidence data, determine the macro-dimensional set corresponding to the domain tags based on a preset global knowledge base, obtain the structured historical data corresponding to the core entities and macro-dimensional set from the preset global knowledge base, generate an environment retrieval expression based on the core entities and macro-dimensional set, and obtain the current hard data based on the environment retrieval expression; The current hard data and structured historical data are combined to obtain macro-environmental data, and weighting factors are generated based on the macro-environmental data. The baseline data and the high-confidence data are corrected according to the weighting factors to obtain the baseline correction result and the high-confidence correction result. The core difference is obtained by subtracting the baseline correction result from the high-confidence correction result.

[0061] The apparatus provided in this disclosure, wherein the deduction module is specifically used for: A macro-environment summary is determined based on the aforementioned macro-environment data; Based on the preset template, the correction results, core gaps and macro-environment summary are assembled to obtain inference prompts; The deduction prompt is input into a preset deduction model, the hypothetical purpose of the deduction prompt is output based on the preset deduction model, and an action path is generated based on the hypothetical purpose, wherein one hypothetical purpose corresponds to at least two action paths; The probability of occurrence of all action paths is calculated based on the preset inference model. The action paths are then sorted in descending order according to the probability of occurrence. The first action path and the hypothetical objective corresponding to the first action path are output to obtain the basis for countermeasures. Based on the preset deduction model, several countermeasure action suggestions are generated according to the countermeasure basis, and the countermeasure action suggestions are deduced to obtain the deduction results; An assessment report is generated by combining the text of the task to be assessed, the correction results, the core gaps, the summary of the macro environment, the basis for countermeasures, and the deduction results.

[0062] The task text analysis device provided in this disclosure can execute a task text analysis method provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects of the execution method.

[0063] To implement the above embodiments, this disclosure also proposes a computer program product, including a computer program / instruction, which, when executed by a processor, implements the task text analysis method described above.

[0064] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure.

[0065] The following is a detailed reference. Figure 3 The diagram illustrates a structural schematic suitable for implementing the electronic device in the embodiments of this disclosure. The electronic device in the embodiments of this disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 3The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0066] like Figure 3 As shown, the electronic device may include a processor (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 302 or a program loaded from memory 308 into random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device. The processor 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0067] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.

[0068] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 309, or installed from a memory 308, or installed from a ROM 302. When the computer program is executed by the processor 301, it performs the functions defined in the task text analysis method of embodiments of this disclosure.

[0069] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0070] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol, such as HTTP (Hypertext Transfer Protocol), and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (LANs), wide area networks (WANs), the Internet (e.g., the Internet), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0071] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0072] The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the aforementioned task text analysis method.

[0073] Electronic devices can be programmed with computer program code in one or more programming languages ​​or combinations thereof to perform the operations of this disclosure. These programming languages ​​include, but are not limited to, object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0074] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0075] The units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0076] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0077] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0078] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0079] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0080] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A method for analyzing task text, characterized in that, include: The text of the task to be analyzed is broken down into several sub-search intentions, and a targeted search expression is generated based on the sub-search intentions and the engine characteristics of the preset search engine. Web page data is obtained based on the targeted search expression. Based on the preset information dehydration end, key sentences in the web page data are identified, and key objective entity indicators are extracted from the key sentences. Supporting data is obtained based on the key objective entity indicators, and the supporting data is subjected to three-dimensional verification to obtain high confidence data. The high confidence data includes core entities and domain tags. Based on the high-confidence data, macro-environmental data is retrieved from a preset global knowledge base. The macro-environmental data is used as a weighting factor to correct the high-confidence data and the preset baseline data. The core gap is calculated based on the correction results. The correction results include the baseline correction results and the high-confidence correction results. Based on the preset simulation model, the task to be assessed is simulated according to the correction results, core gaps and macro-environment data, and an assessment report of the task to be assessed is obtained.

2. The method according to claim 1, characterized in that, Also includes: Store the assessment report in a preset archive; Based on the high confidence correction results in the assessment report, the core entity retrieves the historical baseline data of the core entity from the preset archive, and performs time-series evolution on the historical baseline data to obtain the historical high confidence correction results of the core entity. By comparing the historical high-confidence correction results with the high-confidence correction results, the logical mutation points of the core entity are identified; The logical mutation points are classified, and the weight factors are adjusted based on a preset category weight table.

3. The method according to claim 1, characterized in that, The text of the task to be analyzed is broken down into several sub-search intentions, and targeted search expressions are generated based on the sub-search intentions and the engine characteristics of a preset search engine. Web page data is obtained based on the targeted search expressions, including: Extract entity words from the text of the task to be analyzed, and match key variables from a preset variable library based on the entity words; The key variables are combined to obtain several hypothetical propositions. Based on the hypothetical propositions, the text of the task to be evaluated is decomposed into several sub-retrieval intentions. The number of hypothetical propositions is the same as the number of sub-retrieval intentions. The sub-search intents are effectively filtered based on a preset source credibility classification database to obtain the filtering results; Based on the engine features of the preset search engine and the filtering results, a targeted retrieval expression is generated, and web page data is obtained based on the targeted retrieval expression.

4. The method according to claim 1, characterized in that, Based on preset information, the dehydration process identifies key statements in the webpage data, extracts key objective entity indicators from these statements, obtains supporting data based on these indicators, and performs three-dimensional verification on the supporting data to obtain high-confidence data, including: The web page data is segmented into independent sentence sequences according to punctuation marks, and semantic vectors are generated based on the independent sentence sequences. Based on the preset information dehydration end, the semantic vector is semantically scored to obtain the semantic score corresponding to each semantic vector. The semantic vectors with semantic scores exceeding the preset scoring threshold are stripped to obtain key sentences. Based on the preset indicator table, key objective entity indicators are extracted from the key statements. Based on the engine features, the entities, actions, and values ​​in the key objective entity indicators are decomposed to obtain the verification expression. Based on the verification expression, supporting data is obtained, and supporting data from the same data source are merged according to the preset domain name attribution mapping table to obtain the merged result; The credibility of the data source corresponding to the merged result is evaluated based on a preset source credibility classification library. Data sources with credibility higher than the preset credibility and their corresponding merged results are retained to obtain the retained result. The retained results are validated in three dimensions based on entity, action, and numerical dimensions to obtain high-confidence data.

5. The method according to claim 4, characterized in that, The retained results are validated in three dimensions—entity, action, and value—to obtain high-confidence data, including: Based on a preset entity recognition model, extract the verification subject corresponding to each piece of supporting data in the retained results and the key objective entity indicators in the entity dimension; Extract the action predicates from the supporting data containing the verification subject, determine the semantic direction of the action predicates based on the key objective entity indicators in the action dimension corresponding to the action direction, and filter the supporting data in the retained results according to the semantic direction determination results to obtain the filtered supporting data. If the action predicate of the supporting data is in the same direction as the action dimension in the key objective entity indicator, then the action is deemed reasonable and the supporting data is retained. If the action predicate in the supporting data is opposite to or unrelated to the action dimension in the key objective entity indicator, it is determined to be an action conflict, and the supporting data is removed. Extract the numerical range of the filtered supporting data, determine the range of the numerical range and the corresponding indicator values ​​of the key objective entity indicators in the numerical dimension, obtain the range determination result, and further filter the filtered supporting data based on the range determination result to obtain three-dimensional compliance data. If the range determination result indicates a contradiction in magnitude, the filtered supporting data is removed; if there is no contradiction in magnitude, the filtered supporting data is retained. The number of independent information sources corresponding to the three-dimensional compliance data is counted, and the confidence level of the corresponding independent information source is calculated based on the number of independent information sources and preset rules. Based on the calculation results, high-confidence data in the three-dimensional compliance data is determined.

6. The method according to claim 1, characterized in that, Based on the high-confidence data, macro-environmental data is retrieved from a preset global knowledge base. This macro-environmental data is then used as a weighting factor to correct the high-confidence data against a preset baseline. The core gap is calculated based on the correction result, including: Extract the core entities and domain tags from the high-confidence data, determine the macro-dimensional set corresponding to the domain tags based on a preset global knowledge base, obtain the structured historical data corresponding to the core entities and macro-dimensional set from the preset global knowledge base, generate an environment retrieval expression based on the core entities and macro-dimensional set, and obtain the current hard data based on the environment retrieval expression; The current hard data and structured historical data are combined to obtain macro-environmental data, and weighting factors are generated based on the macro-environmental data. The baseline data and the high-confidence data are corrected according to the weighting factors to obtain the baseline correction result and the high-confidence correction result. The core difference is obtained by subtracting the baseline correction result from the high-confidence correction result.

7. The method according to claim 1, characterized in that, Based on a pre-defined simulation model, the task to be assessed is simulated according to the correction results, core gaps, and macro-environmental data, resulting in an assessment report for the task to be assessed, including: A macro-environment summary is determined based on the aforementioned macro-environment data; Based on the preset template, the correction results, core gaps and macro-environment summary are assembled to obtain inference prompts; The deduction prompt is input into a preset deduction model, the hypothetical purpose of the deduction prompt is output based on the preset deduction model, and an action path is generated based on the hypothetical purpose, wherein one hypothetical purpose corresponds to at least two action paths; The probability of occurrence of all action paths is calculated based on the preset inference model. The action paths are then sorted in descending order according to the probability of occurrence. The first action path and the hypothetical objective corresponding to the first action path are output to obtain the basis for countermeasures. Based on the preset deduction model, several countermeasure action suggestions are generated according to the countermeasure basis, and the countermeasure action suggestions are deduced to obtain the deduction results; An assessment report is generated by combining the text of the task to be assessed, the correction results, the core gaps, the summary of the macro environment, the basis for countermeasures, and the deduction results.

8. A task text analysis device, characterized in that, The device includes: The acquisition module is used to decompose the text of the task to be analyzed into several sub-retrieval intentions, generate targeted retrieval expressions based on the sub-retrieval intentions and the engine features of the preset search engine, and acquire web page data based on the targeted retrieval expressions; The extraction module is used to identify key statements in the web page data based on preset information dehydration end, extract key objective entity indicators from the key statements, obtain supporting data based on the key objective entity indicators, and perform three-dimensional verification on the supporting data to obtain high confidence data. The calculation module is used to retrieve macro-environmental data from a preset global knowledge base based on the high-confidence data, use the macro-environmental data as a weighting factor to correct the high-confidence data and the preset baseline data, and calculate the core gap based on the correction result. The correction result includes the baseline correction result and the high-confidence correction result. The deduction module is used to deduce the task to be assessed based on the correction results, core gaps and macro-environment data, and obtain an assessment report of the task to be assessed.

9. An electronic device, characterized in that, include: Memory; processor; as well as Computer programs; The computer program is stored in the memory and configured to be executed by the processor to implement the steps of the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program / instruction thereon, which, when executed by a processor, implements the steps of the method described in any one of claims 1-7.

Images