WO2023029462A1 - 热点事件状态评估方法 - Google Patents

热点事件状态评估方法 Download PDF

Info

Publication number
WO2023029462A1
WO2023029462A1 PCT/CN2022/083840 CN2022083840W WO2023029462A1 WO 2023029462 A1 WO2023029462 A1 WO 2023029462A1 CN 2022083840 W CN2022083840 W CN 2022083840W WO 2023029462 A1 WO2023029462 A1 WO 2023029462A1
Authority
WO
WIPO (PCT)
Prior art keywords
evaluation
quantitative
index
model
event
Prior art date
Application number
PCT/CN2022/083840
Other languages
English (en)
French (fr)
Inventor
潘磊
代翔
戴礼灿
崔莹
廖泓舟
高翔
陈伟晴
Original Assignee
西南电子技术研究所（中国电子科技集团公司第十研究所）
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 西南电子技术研究所（中国电子科技集团公司第十研究所） filed Critical 西南电子技术研究所（中国电子科技集团公司第十研究所）
Publication of WO2023029462A1 publication Critical patent/WO2023029462A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G06F16/334—Query execution
    • G06F16/3344—Query execution using natural language analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/205—Parsing
    • G06F40/211—Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/01—Social networking

Definitions

  • the invention relates to a hot event analysis technology in the field of information processing, in particular to an event status evaluation method, especially a hot event status evaluation method incorporating expert knowledge.
  • Hotspot events refer to events and issues that arouse widespread social concern or arouse great social repercussions within a certain period of time. An event becomes a hot spot, or gradually formed after a certain period of propagation and brewing, or suddenly erupts due to some repressed social emotional catharsis. The response to emergencies or hotspot events often needs to be carried out under the circumstances of huge risks and limited time, accurately assess the state of the event, and make corresponding decisions in a timely manner, which has high timeliness and accuracy requirements.
  • identification is a relatively simple task.
  • a simple analysis based on the popularity of specific network events can determine whether it is a hotspot event, and the hotspot detection of public opinion events is a specific data requirement.
  • Higher work tasks, so accurate data statistics, calculations and comparative analysis are required to achieve.
  • the biggest feature of public opinion events is their strong variability. According to the development and changes of events in different periods, it is necessary to master synchronous information and data in order to be able to accurately reflect the hotness of events at any time, so as to fully grasp the trend of Internet public opinion.
  • Public opinion risk assessment is an important part of public opinion management and the basis for preventing the rapid spread of negative public opinion. If we can intervene and communicate effectively when public opinion risks appear, we may resolve some potential risks and greater hidden dangers in advance. Public opinion risk assessment is an important part of public opinion guidance. In the new communication environment, public opinion handling needs to move forward, scientifically grasp the law of the generation and evolution of public opinion, and build a more scientific public opinion risk assessment system.
  • the risk assessment model is a relatively mature assessment method, which consists of three parts: risk analysis, risk assessment and risk assessment.
  • Risk analysis includes qualitative risk analysis and quantitative risk analysis.
  • Qualitative risk analysis refers to the determination of probabilities and consequences in a fully qualitative manner.
  • Quantitative risk analysis makes mathematical estimates of probabilities and consequences, taking into account relevant uncertainties.
  • Risk assessment is a process of judging the tolerance of risks based on risk analysis, considering social, economic, environmental and other factors. The whole linking risk analysis and risk evaluation is risk assessment. It can be seen that the risk assessment model system has strong correlation, but it has high complexity, poor portability and low timeliness.
  • the evaluation index system and evaluation model in the current research are complex and diverse, with poor operability and limited scope of application.
  • the multi-hypothesis analysis method is mainly used to complete the status assessment through several steps.
  • the first step is to put forward several hypotheses for the status of the hot events concerned;
  • the second step is to list all relevant indicators;
  • the third step is to make a matrix diagram, put the assumptions obtained in the first step and the indicators obtained in the second step into the matrix diagram, and run horizontally It is a variety of hypotheses, and the vertical columns are evidence and arguments;
  • the fourth step is to make a provisional conclusion on the possibility of each hypothesis;
  • the fifth step is to combine some of the closer hypotheses;
  • the sixth step is to find out that the situation is moving towards the desired The key indicator of development in an unexpected direction;
  • the seventh step experts score based on experience and count the hypothesis with the highest score as the most likely state of the hot event. It can be seen that the multi-hypothesis analysis method is simple, but there are many manual participation, strong dependence on expert knowledge, and low level of intelligence.
  • the present invention aims at the shortcomings of current status evaluation technology of hotspot events, and proposes a hotspot event status evaluation method that incorporates expert knowledge to realize intelligent quantitative estimation of hotspot event status.
  • the purpose of this disclosure is to provide a method that can improve the accuracy of the evaluation model, improve the calculation efficiency, and integrate intelligent An event state assessment method based on expert knowledge.
  • a method for assessing the state of a hotspot event comprising the following steps:
  • Quantitative evaluation model generation According to the important/hot events of concern, complete the construction of multi-level and multi-granularity index systems and evaluation conclusions driven by expert knowledge, covering all dimensions of current hot events of concern; based on the evaluation index system and evaluation conclusions, Combined with the expert knowledge base, complete the matrix filling, matrix updating and matrix quantification of the support degree of the evaluation index to the evaluation conclusion, and generate a quantitative evaluation model;
  • Semantic matching evaluation Based on the quantitative evaluation model, according to the discovery results of hot events, extract the event sentences and speech sentences in the news data of open source hot events as index data, and extract the index data to be quantified from the index database for index type discrimination.
  • the quantification matrix Figure through the matching calculation of the semantic similarity between the index data and the evaluation index, the quantitative weight of the index data is obtained, and the expert knowledge is intelligently integrated into the expert knowledge base to automatically complete the construction of a new matrix, and then the conclusion confidence is calculated, and the quantification corresponding to each conclusion is carried out. Weight statistics and normalized calculations to obtain a preliminary quantitative estimate of the current state of hot events;
  • In-depth classification evaluation After the semi-automatic evaluation matrix reaches a certain number of thresholds, the in-depth classification model training is carried out according to the index data + quantitative labels output by the new matrix, the index data to be quantified is extracted from the index database, and the classification model is used to classify the index data The weight is quantified, and the elements of the matrix diagram are updated to automatically calculate the support degree of the indicator data to the event state, and complete the in-depth automatic quantitative estimation of the event state.
  • this disclosure is driven by expert knowledge to complete the construction of multi-level and multi-granularity index systems and evaluation conclusions, covering all dimensions of current hot events of concern; based on the evaluation index system and evaluation conclusions, combined with expert knowledge library, complete the matrix filling, matrix updating and matrix quantification of the support degree of evaluation indicators to the evaluation conclusions, and generate a quantitative evaluation model.
  • a multi-granularity evaluation index system it covers all dimensions of current hot issues of concern in an all-round way, and improves the evaluation The comprehensiveness of the indicator system.
  • This disclosure is based on the quantitative evaluation model, and according to the discovery results of hot events, event sentences and speech sentences in the news data of open source hot events are extracted as index data, and the index data to be quantified is extracted from the index database for index type discrimination.
  • the quantification matrix diagram Through the matching calculation of the semantic similarity between the index data and the evaluation index, the quantitative weight of the index data is obtained, and the expert knowledge is intelligently integrated into the expert knowledge base to automatically complete the construction of a new matrix, and then the conclusion confidence is calculated, and the quantitative weight statistics corresponding to each conclusion are calculated. With the normalized calculation, the preliminary quantitative estimation of the current hot event status is obtained.
  • the establishment of the event status evaluation conclusion driven by expert knowledge clarifies the direction of the evaluation; the hot event evaluation model is constructed through the evaluation index system and evaluation conclusion, which is effective Improved accuracy for evaluating models.
  • This disclosure combines the empirical knowledge output by the expert knowledge base to calculate the confidence of the conclusion.
  • the deep classification evaluation of the deep neural network classification model is used to automatically complete the calculation of the support degree of the event data to the event state. , to complete the deep automatic quantitative estimation of event status.
  • the quantitative evaluation model uses semantic similarity matching calculation and deep neural network classification to complete the quantitative evaluation of the current hot event state, and integrates expert knowledge in an intelligent way to complete the hot event state estimation, with a high level of intelligence.
  • the evaluation technology proposed in this disclosure is not limited to the evaluation of hot events in a certain field, it can be applied to hot events such as finance, politics, military affairs, security, etc. Analytical decision-maker's right-hand man.
  • Fig. 1 is a block diagram of the hot event state evaluation method of the present invention blended into expert knowledge
  • Figure 2 is a schematic diagram of the construction of the quantitative evaluation model
  • Fig. 3 is a flowchart of the semantic matching evaluation of Fig. 1;
  • FIG. 4 is a flowchart of the deep classification evaluation of FIG. 1 .
  • the method for assessing the status of hotspot events adopts the following steps:
  • Quantitative evaluation model generation and construction According to the important/hot events of concern, the establishment of a hierarchical index system and the construction of evaluation conclusions are driven by expert knowledge, covering all dimensions of current hot events of concern; based on the evaluation index system and evaluation conclusions, combined with experts The estimation conclusions and indicators given by the knowledge base experts and their relationship corrections are completed, the matrix diagram filling, matrix diagram updating and matrix diagram quantification of the evaluation index support degree to the evaluation conclusion are completed, a quantitative evaluation model is generated, and a multi-granularity evaluation index system is constructed;
  • Semantic matching evaluation Based on the quantitative evaluation model, according to the hot event discovery results, extract the event sentences and speech sentences in the open source hot event news data as index data, and extract the quantified index data from the index database for index type discrimination. Graph quantification, dictionary quantification and index category are used to determine the quantified value of the index. Through the semantic similarity matching calculation of the evaluation index, the quantitative weight of the index data is obtained, and the expert knowledge of the expert knowledge base is intelligently integrated and the new matrix is automatically constructed, and then the conclusion is confident. Degree calculation, through the quantitative weight statistics and normalized calculation corresponding to each conclusion, a preliminary quantitative estimate of the current hot event status is obtained;
  • In-depth classification evaluation After the semi-automatic evaluation matrix reaches a certain number of thresholds, the in-depth classification model training is carried out according to the index data + quantitative labels output by the new matrix, the index data to be quantified is extracted from the index database, and the classification model is used to classify the index data Weight quantification, combined with the expert knowledge of the expert knowledge base and quantitative weight, updates the elements of the new matrix, automatically calculates the support degree of the index data to the event state, and uses semantic similarity matching calculation and deep neural network classification to complete the quantitative evaluation of the current hot event state And conclusion confidence calculation, to complete the deep automatic quantitative estimation of event status.
  • the evaluation method of hotspot event status adopts three parts including quantitative evaluation model construction, semantic matching evaluation and in-depth classification evaluation; the quantitative evaluation model construction part establishes a multi-granularity evaluation index system and event The status estimation conclusion, combined with the experience and knowledge of experts, completes the construction of the quantitative evaluation model; the semantic matching evaluation part is based on the quantitative evaluation model, and the preliminary estimation of the state of hot events is completed through the method of semantic similarity matching calculation; the in-depth classification evaluation part is semi-automated After the evaluation matrix reaches a certain number of thresholds, the support calculation of the event data to the event state is realized through the deep neural network classification model, and the deep automatic quantitative estimation of the event state is completed.
  • Quantitative evaluation model construction is divided into three parts: evaluation index construction, evaluation conclusion preset, and quantitative evaluation model generation.
  • evaluation index construction is based on expert experience and knowledge, and typical indicators that meet the index coverage are extracted from event data to meet the index coverage requirements.
  • the comprehensiveness of the evaluation conclusion is based on the user's needs to present the estimated state of the current hot event; in the generation of the quantitative evaluation model, based on the evaluation index and evaluation conclusion, combined with expert knowledge, the evaluation index supports the evaluation conclusion The scoring, and finally generate a quantitative evaluation model.
  • the quantitative evaluation model matches the semantic similarity between the index data and the evaluation index, completes the semantic feature extraction of the index data and evaluation index through the BERT pre-training model, and uses the cosine distance as the similarity measure standard to compare each index
  • the data is matched to the corresponding evaluation index, and the matching result is obtained.
  • the semantic matching model assigns the support degree of the corresponding evaluation index to the evaluation conclusion to the corresponding index data, and generates a semantic matching evaluation matrix.
  • the matching evaluation matrix calculates the statistics and normalization of the support degree of the index data under each evaluation conclusion, and obtains the quantitative estimation of the current hot event status.
  • the deep neural network integrates the data in the semantic matching evaluation matrix according to the label definition to form a deep neural network classification model with labeled training data.
  • the training data is input into the deep neural network classification model after training.
  • the in-depth classification evaluation model automatically completes the calculation of the support degree of the evaluation conclusion by the index data in a classified manner, implements continuous hot data classification and statistical normalization calculation, and obtains a deep quantitative estimate of the current event state .

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Development Economics (AREA)
  • Marketing (AREA)
  • Educational Administration (AREA)
  • Databases & Information Systems (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Tourism & Hospitality (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Game Theory and Decision Science (AREA)
  • Computing Systems (AREA)
  • Primary Health Care (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明公开的一种热点事件状态评估方法，能够改善评估模型精确性，提高计算效率，本发明的热点事件状态评估方法包括：量化评估模型生成：构建多层次多粒度评估指标体系，基于评估指标体系和评估结论，结合专家知识完成评估指标对评估结论支撑度的判定，生成量化评估模型；语义匹配评估：以量化评估模型为基础，通过指标数据与评估指标的匹配得到量化矩阵图，对每个结论对应的量化权重统计与归一化计算，得到当前热点事件状态的初步定量估计；深度分类评估：基于量化矩阵图，利用分类模型进行指标数据分类权重量化，自动计算指标数据对事件状态的支撑度，完成热点事件状态的深度定量评估。

Description

热点事件状态评估方法
本申请要求于2021年08月31日提交中国专利局、申请号为202111011061.4、申请名称“热点事件状态评估方法”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及信息处理领域的热点事件分析技术，具体涉及事件现状评估方法，尤其是融入专家知识的热点事件状态评估方法。
背景技术
当今世界全球化使得人类所处的环境越来越复杂，社会的流动性和复杂性随之迅速增加，现代社会的高速发展导致各个领域和系统的耦合与关联越来越复杂，社会系统的脆弱性也越来越大。在这种背景下，越来越多的不确定因素导致了突发事件、热点事件的发生。热点事件是指在一定时间内出现的引发社会广泛关注或引起社会反响较大的事件和问题。一个事件成为热点，或经一定时期的传播和酝酿逐渐形成，或因某种被压抑的社会情绪宣泄而突然爆发。突发事件或热点事件的应对往往需要在风险巨大且时间有限的情况下进行，准确评估出事件的状态，及时地制定出相应的决策，具有较高的时效性和精准性要求。
互联网的发展改变了人们获取信息的方式，开源互连网络是获取事件数据最直接的来源。大数据舆情分析研判是舆情工作者通过收集分析互联网上关于社会热点或网民关注焦点事件的大量消息报道，发掘背后隐藏关系，进而预测事态发展趋势，为舆情事件处置提供决策参考。网络舆情事件由于主要通过网络传播，而网络也是一个公众性极为突出的平台，因此对于其发生的舆情事件的分析也能够非常容易的通过数据运算实现。网络舆情事件虽然传播于网络，但一般仍然是以现实社会为基础所产生的，因此其现实意义也非常重要，通过对网络上的舆情发生和走向分析，能够从一定程度上反映出社会中的实际舆情走向。微观角度分析网络舆情事件主要就是根据网络上的事件点击、转发、讨论人数等具体信息进行数据的统计、计算和分析，这种分析更加注重实际的网络数据，虽然其结果更加具体化，但由于网络信息量的庞大，其分析工作量也是同样巨大。对网络舆情信息的文档选择和摘要可以通过人工去实现，也可以通过开发的应用程序由机器系统辅助实现。如蚁坊软件的智能化网络舆情分析应用系统中含有自动文档摘要、数据收集分析功能。但是面对海量的网络信息，人们往往只能管中窥豹，不能从整体上把握网络的热点；另一方面，对于同一热点事件的描述，由于信息发布者的角度和立场的不同，网络信息报道方式往往各不相同，信息真实性和全面 性也参差不齐。这对事件数据获取的全面性和评估模型的鲁棒性提出了考验。
在网络舆情事件的识别与检测中，识别是一项相对比较简单的工作，根据具体的网络事件热度进行简单分析就能够判定出是否热点事件，而舆情事件的热点检测是一项对具体数据要求较高的工作任务，因此需要有准确的数据统计、计算和对比分析才能实现。另外舆情事件最大的特点就是变化性强，根据不同时期事件的发展变化走向，需要进行同步信息和数据的掌握，才能够随时准确的反映事件的热点程度，从而全面的掌握网络舆情走向。同时根据事件的多面性也要针对不同的事件反映或讨论结果进行多元化信息分析，对比结果或讨论观点的热度走向等，进而做出相应的舆论干预策略。舆情风险评估是舆情管理的重要组成部分，也是防范负面舆论快速扩散的基础，如果能在舆情风险出现苗头时就进行干预和有效沟通，就可能提前化解某些潜在风险和更大的隐患。舆情风险评估是开展舆情引导的重要环节。在新传播环境下，开展舆情处置需要关口前移，科学把握舆论的生成和演化规律，构建更加科学的舆情风险评估体系。风险评估模型是较为成熟的一种评估方法，由风险分析、风险评价和风险评估三部分组成。风险分析包括定性风险分析和定量风险分析。定性风险分析是指以完全定性的方法确定概率和后果。定量风险分析对概率及后果进行数学估算，需考虑相关的不确定性因素。风险评价是在风险分析的基础上，考虑社会、经济、环境等方面的因素，对风险的容忍度做出判断过程。将风险分析和风险评价连接起来的整体即为风险评估。可以看出风险评估模型系统关联性强，但复杂度高、可移植性差、时效性低。目前研究中的评估指标体系和评估模型复杂多样，操作性较差，适用范围有限。
目前采用多假设分析方法主要是通过几个步骤完成状态评估。第一步先提出针对关注热点事件的状态的几种假设；第二步列举出所有相关指标；第三步制作矩阵图，把第一步得到假设和第二步得到指标放入矩阵图，横行是各种假设，竖列是证据和论据；第四步对每种假设的可能性做出临时结论；第五步把某些比较接近的假设予以合并；第六步找出表明事态正朝着意想不到的方向发展的关键指标；第七步专家根据经验打分后统计得出得分最高的假设，作为热点事件的最可能状态。可以看出多假设分析方法简单，但是人工参与较多、对专家知识依赖强，智能化水平不高。
综上所述，本发明针对目前热点事件现状评估技术存在的短板弱项，提出了一种融入专家知识的热点事件状态评估方法，实现关注热点事件状态的智能化定量估计。
发明内容
本公开的目的是针对目前事件状态估计方法存在的人工参与度高、智能化水平低、主 观偏差大、计算效率慢等问题，提供一种能够改善评估模型精确性，提高计算效率，智能化融入专家知识的事件状态评估方法。
本公开的上述目的可以通过以下技术方案予以实现：一种热点事件状态评估方法，包括如下步骤：
量化评估模型生成：根据关注的重/热点事件，以专家知识为驱动完成多层次多粒度指标体系构建和评估结论构建，全方位覆盖当前关注热点事件的各个维度；基于评估指标体系和评估结论，结合专家知识库，完成评估指标对评估结论支撑度的矩阵图填充、矩阵图更新和矩阵图量化，生成量化评估模型；
语义匹配评估：以量化评估模型为基础，根据热点事件发现结果，提取开源热点事件新闻数据中的事件句、言论句作为指标数据，从指标数据库提取待量化指标数据进行指标类型判别，根据量化矩阵图，通过指标数据与评估指标的语义相似度匹配计算，得到指标数据的量化权重，智能融入专家知识库专家知识自动完成新矩阵图构建，然后进行结论置信度计算，通过每个结论对应的量化权重统计与归一化计算，得到当前热点事件状态的初步定量估计；
深度分类评估：在半自动化评估矩阵图达到一定数量的阈值后，根据新矩阵图输出的指标数据+量化标签进行深度分类模型训练，从指标数据库提取待量化指标数据，利用分类模型进行指标数据分类权重量化，并更新矩阵图元素，自动计算指标数据对事件状态的支撑度，完成事件状态的深度自动化定量估计。
本公开相比于现有技术具有如下有益效果：
本公开根据关注的重/热点事件，以专家知识为驱动完成多层次多粒度指标体系构建和评估结论构建，全方位覆盖当前关注热点事件的各个维度；基于评估指标体系和评估结论，结合专家知识库，完成评估指标对评估结论支撑度的矩阵图填充、矩阵图更新和矩阵图量化，生成量化评估模型，通过多粒度评估指标体系构建，全方位覆盖当前关注热点事件的各个维度，提升了评估指标体系的全面性。本公开以量化评估模型为基础，根据热点事件发现结果，提取开源热点事件新闻数据中的事件句、言论句作为指标数据，从指标数据库提取待量化指标数据进行指标类型判别，根据量化矩阵图，通过指标数据与评估指标的语义相似度匹配计算，得到指标数据的量化权重，智能融入专家知识库专家知识自动完成新矩阵图构建，然后进行结论置信度计算，通过每个结论对应的量化权重统计与归一化计算，得到当前热点事件状态的初步定量估计，这种以专家知识驱动事件状态评估结论的建立，明确了评估的方向性；通过评估指标体系和评估结论构建热点事件评估模型，有效改善了评估模型的精确性。
本公开结合专家知识库输出的经验知识，进行结论置信度计算，在半自动化评估矩阵达到一定数量的阈值后，采用深度神经网络分类模型深度分类评估，自动完成事件数据对事件状态的支撑度计算，完成深度自动化定量估计事件状态。这种以量化评估模型为基础，采用语义相似度匹配计算和深度神经网络分类完成当前热点事件状态的定量评估，通过智能化方式融入专家知识完成热点事件状态估计，智能化水平高。
本公开提出的评估技术不局限于某一领域的热点事件评估，其可适用于金融、政外、军事、安全等类型热点事件，具有很好的普适性和实用性，是各行各业的分析决策人员的得力帮手。
附图说明
图1是本发明融入专家知识的热点事件状态评估方法流程框图；
图2是量化评估模型构建示意图；
图3是图1的语义匹配评估的流程图；
图4是图1的深度分类评估的流程图。
具体实施方式
参阅图1。根据本发明，热点事件状态评估方法采用如下步骤：
量化评估模型生成构建：根据关注的重/热点事件，以专家知识为驱动完成层次指标体系构建和评估结论构建，全方位覆盖当前关注热点事件的各个维度；基于评估指标体系和评估结论，结合专家知识库专家所给出估计结论和指标及其两者关系修正，完成评估指标对评估结论支撑度的矩阵图填充、矩阵图更新和矩阵图量化，生成量化评估模型，构建多粒度评估指标体系；
语义匹配评估：以量化评估模型为基础，根据热点事件发现结果，提取开源热点事件新闻数据中的事件句、言论句作为指标数据，从指标数据库提取待量化指标数据进行指标类型判别，根据来自矩阵图量化、字典量化和指标类别进行指标量化值确定，通过评估指标的语义相似度匹配计算，得到指标数据的量化权重，智能融入专家知识库专家知识并自动完成新矩阵图构建，然后进行结论置信度计算，通过每个结论对应的量化权重统计与归一化计算，得到当前热点事件状态的初步定量估计；
深度分类评估：在半自动化评估矩阵图达到一定数量的阈值后，根据新矩阵图输出的指标数据+量化标签进行深度分类模型训练，从指标数据库提取待量化指标数据，利用分类模型进行指标数据分类权重量化，结合专家知识库专家知识和量化权重，对新矩阵图元素更新，自动 计算指标数据对事件状态的支撑度，采用语义相似度匹配计算和深度神经网络分类完成当前热点事件状态的定量评估和结论置信度计算，完成事件状态的深度自动化定量估计。
在可选的实施例中，热点事件状态评估方法采用包括量化评估模型构建、语义匹配评估和深度分类评估三部分；量化评估模型构建部分基于热点事件发现的事件数据建立多粒度评估指标体系和事件状态预估结论，结合融入专家经验知识，完成量化评估模型构建；语义匹配评估部分以量化评估模型为基础，通过语义相似度匹配计算的方法完成热点事件状态初步估计；深度分类评估部分在半自动化评估矩阵达到一定数量的阈值后，通过深度神经网络分类模型实现事件数据对事件状态的支撑度计算，完成深度自动化定量估计事件状态。
参见图2。量化评估模型构建分为评估指标构建、评估结论预置和量化评估模型生成三部分，其中，评估指标构建根据专家经验知识，从事件数据中提炼出满足指标覆盖的典型性指标，以满足指标覆盖的全面性；评估结论预置部分则依据用户需求对当前热点事件估计的状态进行文字化呈现；在量化评估模型生成中，基于评估指标和评估结论，结合专家知识完成评估指标对评估结论支撑度的打分，最终生成量化评估模型。
参见图3。在语义匹配评估阶段，量化评估模型将指标数据与评估指标进行语义相似度匹配，通过BERT预训练模型完成指标数据和评估指标的语义特征提取，以余弦距离作为相似度衡量标准，将每一条指标数据匹配到相应的评估指标，获得匹配结果。语义匹配模型将相应评估指标对评估结论的支撑度赋值给对应的指标数据，生成语义匹配评估矩阵。匹配评估矩阵通过计算每条评估结论下指标数据对其支撑度的统计和归一化后，得到当前热点事件状态的定量估计。
参见图4。在深度分类评估中，深度神经网络根据标签定义将语义匹配评估矩阵中的数据进行整合，形成带有标签训练数据的深度神经网络分类模型，同时将训练数据输入深度神经网络分类模型训练完成后，生成深度分类评估矩阵。深度分类评估模型在当新的指标数据输入时，自动以分类方式完成指标数据对评估结论支撑度的计算，实施持续性的热点数据分类和统计归一化计算，得到当前事件状态的深度定量估计。
以上所述为本发明较佳实施例，应该注意的是上述实施例对本发明进行说明，然而本发明并不局限于此，并且本领域技术人员在脱离所附权利要求的范围情况下可设计出替换实施例。对于本领域内的普通技术人员而言，在不脱离本发明的精神和实质的情况下，可以做出各种变型和改进，这些变型和改进也视为本发明的保护范围。

Claims (8)

  1. 一种热点事件状态评估方法，包括如下步骤：
    量化评估模型生成：根据关注的重/热点事件，以专家知识为驱动完成多层次多粒度指标体系构建和评估结论构建，全方位覆盖当前关注热点事件的各个维度；基于评估指标体系和评估结论，结合专家知识库，完成评估指标对评估结论支撑度的矩阵图填充、矩阵图更新和矩阵图量化，生成量化评估模型；
    语义匹配评估：以量化评估模型为基础，根据热点事件发现结果，提取开源热点事件新闻数据中的事件句、言论句作为指标数据，从指标数据库提取待量化指标数据进行指标类型判别，根据量化矩阵图，通过指标数据与评估指标的语义相似度匹配计算，得到指标数据的量化权重，智能融入专家知识库专家知识自动完成新矩阵图构建，然后进行结论置信度计算，通过每个结论对应的量化权重统计与归一化计算，得到当前热点事件状态的初步定量估计；
    深度分类评估：在半自动化评估矩阵图达到一定数量的阈值后，根据新矩阵图输出的指标数据+量化标签进行深度分类模型训练，从指标数据库提取待量化指标数据，利用分类模型进行指标数据分类权重量化，并更新矩阵图元素，自动计算指标数据对事件状态的支撑度，完成事件状态的深度自动化定量估计。
  2. 如权利要求1所述的热点事件状态评估方法，其中：热点事件状态评估方法包括量化评估模型构建、语义匹配评估和深度分类评估三部分；量化评估模型构建部分基于热点事件发现的事件数据建立多粒度评估指标体系和事件状态预估结论，智能融入专家经验知识，完成量化评估模型构建；语义匹配评估部分以量化评估模型为基础，通过语义相似度匹配计算的方法完成热点事件状态初步估计；深度分类评估部分在半自动化评估矩阵达到一定数量的阈值后，通过深度神经网络分类模型实现事件数据对事件状态的支撑度计算，完成深度自动化定量估计事件状态。
  3. 如权利要求1或2所述的热点事件状态评估方法，其中：量化评估模型构建分为评估指标构建、评估结论预置和量化评估模型生成三部分，其中，评估指标构建根据专家经验知识，从事件数据中提炼出满足指标覆盖的典型性指标，以满足指标覆盖的全面性；评估结论预置部分则依据用户需求对当前热点事件估计的状态进行文字化呈现；在量化评估模型生成中，基于评估指标和评估结论，结合专家知识完成评估指标对评估结论支撑度的打分，最终生成量化评估模型。
  4. 如权利要求3所述的热点事件状态评估方法，其中：在语义匹配评估阶段，量化评估模型将指标数据与评估指标进行语义相似度匹配，通过BERT预训练模型完成指标数据和评估指 标的语义特征提取，以余弦距离作为相似度衡量标准，将每一条指标数据匹配到相应的评估指标，获得匹配结果。
  5. 如权利要求4所述的热点事件状态评估方法，其中：语义匹配模型将相应评估指标对评估结论的支撑度赋值给对应的指标数据，生成语义匹配评估矩阵。
  6. 如权利要求5所述的热点事件状态评估方法，其中：匹配评估矩阵通过计算每条评估结论下指标数据对其支撑度的统计和归一化后，得到当前热点事件状态的定量估计。
  7. 如权利要求1所述的热点事件状态评估方法，其中：在深度分类评估中，深度神经网络根据标签定义将语义匹配评估矩阵中的数据进行整合，形成带有标签训练数据的深度神经网络分类模型，同时将训练数据输入深度神经网络分类模型训练完成后，生成深度分类评估矩阵。
  8. 如权利要求7所述的热点事件状态评估方法，其中：深度分类评估模型在当新的指标数据输入时，自动以分类方式完成指标数据对评估结论支撑度的计算，实施持续性的热点数据分类和统计归一化计算，得到当前事件状态的深度定量估计。
PCT/CN2022/083840 2021-08-31 2022-03-29 热点事件状态评估方法 WO2023029462A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111011061.4A CN113779195B (zh) 2021-08-31 2021-08-31 热点事件状态评估方法
CN202111011061.4 2021-08-31

Publications (1)

Publication Number Publication Date
WO2023029462A1 true WO2023029462A1 (zh) 2023-03-09

Family

ID=78840277

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/083840 WO2023029462A1 (zh) 2021-08-31 2022-03-29 热点事件状态评估方法

Country Status (2)

Country Link
CN (1) CN113779195B (zh)
WO (1) WO2023029462A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116523566A (zh) * 2023-06-30 2023-08-01 和元达信息科技有限公司 一种基于互联网广告投放的伪热度识别方法及系统
CN116757314A (zh) * 2023-06-02 2023-09-15 中国人民解放军军事科学院国防科技创新研究院 基于大数据和人工智能融合驱动的技术预见方法和系统
CN117494897A (zh) * 2023-11-14 2024-02-02 西安康奈网络科技有限公司 一种单一舆情事件发展倾向性判断方法
CN117763140A (zh) * 2024-02-22 2024-03-26 神州医疗科技股份有限公司 基于计算特征网络的精准医学信息结论生成方法

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113779195B (zh) * 2021-08-31 2023-12-22 西南电子技术研究所（中国电子科技集团公司第十研究所） 热点事件状态评估方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103049532A (zh) * 2012-12-21 2013-04-17 东莞中国科学院云计算产业技术创新与育成中心 基于突发事件应急管理的知识库引擎构建及其查询方法
US20180124091A1 (en) * 2016-10-27 2018-05-03 Src, Inc. Method for the Continuous Calculation of a Cyber Security Risk Index
CN110633373A (zh) * 2018-06-20 2019-12-31 上海财经大学 一种基于知识图谱和深度学习的汽车舆情分析方法
CN112711691A (zh) * 2021-01-08 2021-04-27 深圳市网联安瑞网络科技有限公司 网络舆情引导效果数据信息处理方法、系统、终端及介质
CN113779195A (zh) * 2021-08-31 2021-12-10 西南电子技术研究所（中国电子科技集团公司第十研究所） 热点事件状态评估方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112765442A (zh) * 2018-06-25 2021-05-07 中译语通科技股份有限公司 基于新闻大数据的网络情绪波动指数监测分析方法及系统
CN110866126A (zh) * 2019-11-22 2020-03-06 福建工程学院 一种高校网络舆情风险评估方法
CN111753093A (zh) * 2020-07-02 2020-10-09 东北电力大学 一种网络舆情危机等级评价方法和装置
CN112016858A (zh) * 2020-11-02 2020-12-01 国网浙江省电力有限公司 一种基于主客观赋权法的舆情风险评价方法
CN112734154B (zh) * 2020-11-16 2023-08-01 中山大学 一种基于模糊数相似度的多因素舆情风险评估方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103049532A (zh) * 2012-12-21 2013-04-17 东莞中国科学院云计算产业技术创新与育成中心 基于突发事件应急管理的知识库引擎构建及其查询方法
US20180124091A1 (en) * 2016-10-27 2018-05-03 Src, Inc. Method for the Continuous Calculation of a Cyber Security Risk Index
CN110633373A (zh) * 2018-06-20 2019-12-31 上海财经大学 一种基于知识图谱和深度学习的汽车舆情分析方法
CN112711691A (zh) * 2021-01-08 2021-04-27 深圳市网联安瑞网络科技有限公司 网络舆情引导效果数据信息处理方法、系统、终端及介质
CN113779195A (zh) * 2021-08-31 2021-12-10 西南电子技术研究所（中国电子科技集团公司第十研究所） 热点事件状态评估方法

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116757314A (zh) * 2023-06-02 2023-09-15 中国人民解放军军事科学院国防科技创新研究院 基于大数据和人工智能融合驱动的技术预见方法和系统
CN116757314B (zh) * 2023-06-02 2024-03-19 中国人民解放军军事科学院国防科技创新研究院 基于大数据和人工智能融合驱动的技术预见方法和系统
CN116523566A (zh) * 2023-06-30 2023-08-01 和元达信息科技有限公司 一种基于互联网广告投放的伪热度识别方法及系统
CN116523566B (zh) * 2023-06-30 2024-01-02 和元达信息科技有限公司 一种基于互联网广告投放的伪热度识别方法及系统
CN117494897A (zh) * 2023-11-14 2024-02-02 西安康奈网络科技有限公司 一种单一舆情事件发展倾向性判断方法
CN117494897B (zh) * 2023-11-14 2024-05-17 西安康奈网络科技有限公司 一种单一舆情事件发展倾向性判断方法
CN117763140A (zh) * 2024-02-22 2024-03-26 神州医疗科技股份有限公司 基于计算特征网络的精准医学信息结论生成方法
CN117763140B (zh) * 2024-02-22 2024-05-28 神州医疗科技股份有限公司 基于计算特征网络的精准医学信息结论生成方法

Also Published As

Publication number Publication date
CN113779195B (zh) 2023-12-22
CN113779195A (zh) 2021-12-10

Similar Documents

Publication Publication Date Title
WO2023029462A1 (zh) 热点事件状态评估方法
CN110796470B (zh) 一种面向市场主体监管和服务的数据分析系统
Wei TODIM method for picture fuzzy multiple attribute decision making
CN106529804B (zh) 基于文本挖掘技术的客户投诉预警监测分析方法
CN104820629B (zh) 一种智能的舆情突发事件应急处理系统及方法
Kou et al. An integrated expert system for fast disaster assessment
CN107220353A (zh) 一种智能客服机器人满意度自动评价方法及系统
CN110675023B (zh) 一种基于神经网络的诉讼请求合理性预测模型训练方法和预测方法及装置
CN104809188A (zh) 一种企业人才流失的数据挖掘分析方法及装置
CN104572709A (zh) 用于企业创新体系的数据挖掘系统
CN107844914B (zh) 基于集团管理的风险管控系统和实现方法
CN109241068A (zh) 前后台数据比对的方法、装置及终端设备
CN115309913A (zh) 一种基于深度学习的财务数据风险识别方法及系统
CN112508743B (zh) 技术转移办公室通用信息交互方法、终端及介质
Liu et al. Research and citation analysis of data mining technology based on Bayes algorithm
Aghimien et al. A review of the application of data mining for sustainable construction in Nigeria
CN105512224A (zh) 基于光标位置序列的搜索引擎用户满意度自动评估方法
CN112132368A (zh) 信息处理方法以及装置、计算设备、存储介质
CN107798137B (zh) 一种基于可加模型的多源异构数据融合架构系统
CN113674846A (zh) 基于lstm网络的医院智慧服务舆情监控平台
Hu Big data analysis of criminal investigations
CN113254512A (zh) 一种军民融合政策信息数据分析优化系统
Shino et al. Implementation of Data Mining with Naive Bayes Algorithm for Eligibility Classification of Basic Food Aid Recipients
CN114529383B (zh) 实现税务缴纳追踪和税款流失预警方法及系统
Qiao et al. Constructing a data warehouse based decision support platform for China tourism industry

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22862609

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

WA Withdrawal of international application