Strategic problem quantification apparatus based on computer natural language processing

Abstract

The application discloses a kind of strategic problem quantitative evaluation devices based on computer natural language processing, belong to computer natural language processing field, comprising: model construction unit creates the strategic problem model to be analyzed, constructs strategic problem index system tree, and the index type, index keyword phrase and index matching rule information are entered for each leaf node of index system tree;Leaf node score calculation unit adopts different leaf node score calculation mode according to index type;Weight setting unit sets the weight of each node in index system tree, and the score of root node is obtained based on leaf node score layer-by-layer recursive weighted summation;Evaluation unit introduces periodic evaluation mode, and according to the actual development situation of problem, index system is adjusted in different periods in time, and the keyword phrase and matching rule of leaf node are modified, and score is recalculated, and the long-term change trend of strategic problem is monitored.The application provides a new device for strategic problem quantitative evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer natural language processing, and more particularly to a strategic problem quantitative evaluation device based on computer natural language processing. BACKGROUND

[0002] The quantitative evaluation of strategic problems aims to assist intelligence analysts in deepening their understanding of the development trend of major strategic problems. Its main research method is expert model decision, which includes three steps: strategic problem modeling, underlying index scoring, and quantitative calculation of evaluation results.

[0003] Strategic problem modeling is mainly constructed by experts in the field based on experience knowledge to build a model framework of the strategic problem to be analyzed, which converts abstract analysis tasks into more specific and detailed index items. The form of the model is generally a tree structure index system. The index system tree describes the connotation of the strategic problem from different subfields and different abstraction levels.

[0004] Underlying index scoring is the process of calculating the score for the leaf nodes of the tree-shaped index system. Generally, it is divided into data-driven and experience-driven methods. The former often needs to collect objective data from official or authoritative think tanks, and uses normalization algorithms or gradient rules to convert data into scores within 100. The latter is independently scored by an expert group based on experience, and the weighted average is obtained.

[0005] Quantitative calculation of evaluation results is the process of calculating the total score of the root node based on the leaf node. The score of each layer node is obtained by averaging the scores of all nodes in the lower layer. Based on the leaf node score, the score of each node in the upper layer can be calculated, and so on, layer by layer, until the quantitative result of the strategic problem root node is obtained.

[0006] The existing strategic problem quantitative evaluation method has several problems:

[0007] 1) The underlying index score relies too much on expert experience, consuming a lot of manpower and resources. The scoring process is highly subjective and ignores the support of objective information materials. It does not use semantic mining and information extraction techniques of computer natural language processing. Semantic mining technology can quantify the amount of index semantic connotation information contained in information materials through the matching of high-dimensional features of text and index vectors. Information extraction is a basic natural language processing algorithm that can convert semi-structured text information into structured entity and entity relationship data. Through pattern matching of index rules and structured data, it can automatically provide exact and interpretable basis for scoring the index;

[0008] 2) The bottom layer indicators are not distinguished. In actual scenarios, different indicators have different contributions to the total score, some indicators have greater impact on the strategic trend, and some have smaller impact; even some indicators will immediately upgrade the strategic problem situation once they occur, and the total score will be greater than a certain threshold.

[0009] 3) The long-term nature of strategic problem evaluation is not considered, and in the long-term evaluation process, the model indicators need to be adjusted according to the actual development situation, or even remodeled. There is no index weight automatic optimization evaluation technology using computer natural language processing, which can supervise the weight prediction model based on the correlation representation of historical index weight and material vector, so as to predict the best index weight distribution according to the new cycle material vector. SUMMARY

[0010] The purpose of the present application is to overcome the shortcomings of the prior art, and to provide a strategic problem quantitative evaluation device based on computer natural language processing, which uses computer automatic score calculation to replace expert scoring, and provides a new device for strategic problem quantitative evaluation, solving the problem that the existing strategic problem quantitative evaluation method mainly relies on manual analysis of indicators and mainly in the field of social science research, and not fully utilizing computer natural language processing technology.

[0011] The purpose of the present application is achieved by the following scheme:

[0012] A strategic problem quantitative evaluation device based on computer natural language processing, comprising a model construction unit, a leaf node score calculation unit, a weight setting unit and an evaluation unit;

[0013] The model construction unit creates a strategic problem model to be analyzed, constructs a strategic problem indicator system tree, and inputs indicator type, indicator keyword phrase and indicator matching rule information for each leaf node of the indicator system tree;

[0014] The leaf node score calculation unit adopts different leaf node score calculation methods according to the indicator type;

[0015] The weight setting unit sets the weight of each node in the indicator system tree, and recursively sums up the weights of each layer based on the leaf node score to obtain the score of the root node, i.e. the strategic problem quantitative evaluation value;

[0016] The evaluation unit introduces a periodic evaluation method, adjusts the indicator system and modifies the keyword phrase and matching rule of the leaf node according to the actual development of the problem in different periods, recalculates the score, and completes the monitoring of the long-term change trend of the strategic problem.

[0017] Further, the indicator type includes "general type indicator-keyword", "general type indicator-rule matching" and "trigger type indicator".

[0018] Further, the different leaf node score calculation methods are adopted according to the index types, and specifically include:

[0019] The 'general type index-keyword' is calculated according to the 'word frequency-document frequency' of the corresponding keyword group in the relevant material text;

[0020] The 'general type index-rule matching' is calculated according to the matching degree of the extracted structured data and the scoring rule;

[0021] The 'trigger type index' is calculated according to whether the intervention to the overall quantitative result occurs.

[0022] Further, in the process of calculating the 'general type index-keyword', the 'general type index-keyword' enters the keyword group, and the keyword group is a description of the semantic connotation of the index node; the logical operator is introduced between the keywords as the basis for automatic scoring.

[0023] Further, in the process of calculating the 'general type index-rule matching', the rule of each leaf node is freely set; the 'general type index-rule matching' enters the scoring formula, substitutes the structured data of the fixed variable, and automatically calculates the score.

[0024] Further, the 'trigger type index' is configured with a state bit and a basic value whether the state bit occurs; once the state bit flag occurs, the basic value will directly participate in the calculation of the root node total score.

[0025] Further, the weight setting unit sets a weight value for each node, and the value range is 0 to 1, which is used to distinguish the different contribution degrees of different index items in the total score calculation.

[0026] Further, the evaluation unit introduces a periodic evaluation method, converts a long-term evaluation strategic problem into a large number of stage summaries, can macroscopically grasp the development context, and can find the key points with large change amplitudes in the development process.

[0027] Further, the processor and the memory are further included, the computer program is stored in the memory, the computer program is loaded and run by the processor, the module architecture of the running model construction unit, the leaf node score calculation unit, the weight setting unit and the evaluation unit is completed, and the long-term change trend of the strategic problem is monitored.

[0028] Further, the display is further included, and the long-term change trend of the strategic problem is presented on the display.

[0029] The beneficial effects of the present application include:

[0030] (1) The device of the present application adopts computer automatic score calculation to replace expert scoring. When constructing the model index system, key word groups are added to the leaf node index, the frequency of the appearance of these key word groups in a large amount of information materials and the number of hit materials are counted, and the weighted term frequency-document frequency is converted into a normalized index score. Expert experience is converted into semantic mining based on massive information. The higher the weighted term frequency-document frequency value is, the higher the heat of the index node in historical materials is, the more influential it is, and the greater its contribution to the total score is. For other special indexes, their semantic connotation can be mapped to the number of specific high-value strategic goals. The named entity recognition technology is used to automatically extract the state or number of special objects, and the score is automatically calculated after matching the judgment rules.

[0031] (2) The device of the present application introduces index types and node weights, and adopts computer automatic score calculation to replace expert scoring. In order to distinguish the contribution of different indexes to the total score, the indexes are divided into "general type index-key word", "general type index-rule matching" and "trigger type index". Each type of index mines clues from a large amount of information materials and automatically completes the calculation. When constructing the strategic problem index system tree, key word groups, pattern matching rules and trigger basic scores and the like information are respectively input for the three types of indexes. The "general type index-key word" index counts the frequency of the appearance of its key word groups in information materials and the number of hit materials, and converts the weighted term frequency-document frequency into a normalized index score. Expert experience is converted into semantic mining based on massive information. The higher the weighted term frequency-document frequency value is, the higher the heat of the index node in historical materials is, the more influential it is, and the greater its contribution to the total score is. The "general type index-rule matching" calculates the score according to the matching degree of the extracted structured data and the scoring rules, and the scoring rules are clear and explicit without human subjective influence. The trigger type index refers to the leaf node that has a strong influence on the total score. Once the situation described by the index occurs, the strategic problem situation will immediately upgrade or decline, and the corresponding total score will directly change.

[0032] (3) The device of the present application adopts a periodic evaluation method to track the long-term development trend of the strategic problem. The index system, index key word groups and node weight values corresponding to each period can be modified to adapt to changes in the objects of attention, different focuses and changes in the connotation of attention in different periods of the strategic problem. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0034] Figure 1 Setting attribute of leaf node for strategic problem model construction;

[0035] Figure 2 Setting weight of strategic topic index system tree;

[0036] Figure 3 Setting trend chart of strategic problem periodic evaluation result. DETAILED DESCRIPTION

[0037] All features disclosed in the embodiments disclosed in the specification, or all steps in the methods or processes impliedly disclosed, can be combined and / or extended, replaced, except for mutually exclusive features and / or steps, in any manner.

[0038] According to the device of the application, firstly, a strategic problem model to be analyzed is created in the device provided by the application, a strategic problem index system tree is constructed, and information such as index type, index keyword phrase, and index matching rule is recorded for each leaf node of the index system tree; secondly, different leaf node score calculation methods are used according to the index type, the score of “general type index-keyword” is calculated according to the “word frequency-document frequency” of the corresponding keyword phrase in the relevant material text, the score of “general type index-rule matching” is calculated according to the matching degree of the extracted structured data and the scoring rule, and the score of “trigger type index” is calculated according to whether it interferes with the overall quantitative result; then, the weight of each node in the index system tree is set, the score of the root node, i.e., the strategic problem quantitative evaluation value, is obtained by layer-by-layer recursive weighted summation based on the leaf node score; finally, a periodic evaluation method is introduced, the index system is adjusted in different periods according to the actual development of the problem, the keyword phrase and matching rule of the leaf node are modified, the score is recalculated, and the monitoring of the long-term change trend of the strategic problem is completed.

[0039] As shown in Figure 1 , the strategic problem modeling process of the device of the application is as follows:

[0040] A model is built for the strategic problem to be analyzed, and the problem of "long-term investment security risk level analysis of a country" is taken as an example. A corresponding index system tree is created in the computer device provided by the device. The index system tree has four layers and contains 27 nodes. The root node is the target of this quantitative evaluation, "long-term investment security risk level analysis of a country". All leaf nodes are provided with index types, including "general index-keyword", "general index-rule matching", and "trigger index". Each type corresponds to a different quantitative method. The "general index-keyword" enters the keyword group. The keyword group is a description of the semantic connotation of the index node and is the basis for subsequent automatic scoring. The "general index-rule matching" enters the scoring formula, and the score is automatically calculated by substituting the structured data of the fixed variable. The "trigger index" is configured with a state bit and a basic value. Once the state bit flag is set to occur, the basic value will directly participate in the calculation of the total score of the root node.

[0041] In the device provided by the device, the score calculation methods of the three index types are introduced below.

[0042] The "general index-keyword" is obtained by calculating the term frequency-inverse document frequency of the keyword group. Assuming that there are n "general index-keyword" indexes, T information materials, and the keyword group of each index is g i (i=1, 2,..,n), the number of information materials hit by each index item is w i (i=1, 2,..,n), then the term frequency-inverse document frequency of each index item is The calculation formula is as follows:

[0043]

[0044] Where k is the weighting coefficient, generally greater than 1, in order to avoid the situation that the score is very low due to the small number of hit documents. represents the frequency of the keyword group, represents the maximum value of the term frequency, which is used for normalization. When the calculation result is greater than 1, the value is uniformly taken as 1.

[0045] Unlike ordinary word count, the keyword group term frequency calculation method contains part of the semantic information. In order to make the keyword group reflect the semantic connotation behind the index item, logical operators are introduced between keywords. Further expansion, the keywords in the first condition are connected by or logical symbols, which means that the inclusion of any one of the keywords in the material is considered to meet the condition. By analogy, the term frequency of the index item can be obtained. Such a calculation method can not only filter a large number of irrelevant texts but also match materials with different expressions and similar semantics.

[0046] "General type index-rule matching" calculates the score by matching the structured data extracted from the information material with the rules. The rules of each leaf node can be freely set. Taking the node "number of annual external capital management acts" as an example, according to previous cases, there is a great correlation between the investment environment of a state-owned enterprise and the number of relevant policies and regulations introduced by the state-owned enterprise in the same year. According to statistics, when the number of foreign capital management policies and regulations exceeds a certain threshold, more than 50% of transnational companies will reduce their business investment in the state-owned enterprise. Therefore, the score rule can be set as:

[0047]

[0048] The number of policies and regulations can be obtained by named entity extraction and relationship extraction of the information material.

[0049] "Trigger type index" sets a basic value, and when the index is marked as occurring, it will directly participate in the total score calculation. Assuming that there are m "trigger type indexes", and the basic values of the m "trigger type indexes" are b i (i=1, 2,..., m). And the total score of other types of index nodes is Z other . Then the total score of the root node of the index system Z total is calculated according to the following formula:

[0050]

[0051] It can be seen that the "trigger type index" can directly affect the value range of the quantitative evaluation result of the strategic problem.

[0052] As shown in Figure 2 , the total score calculation based on weighted summation

[0053] Based on the evaluation values of the leaf nodes of the index system tree obtained by the above three methods, the device sets a weight value (the value range is 0 to 1) for each node to distinguish the different contribution degrees of different index items in the total score calculation. Assuming that the root node has m branches, the score corresponding to each branch node is Z i (i=1, 2,..., m), and the weight value is q i (i=1, 2,..., m), then the root node score calculation formula is:

[0054]

[0055] The calculation method of the score Z i corresponding to the branch node is the same as that of the root node, which depends on the values of the next level nodes. In this way, the calculation is recursively performed until the leaf nodes are reached.

[0056] As shown in Figure 3As shown, the device of the application introduces a periodic evaluation method, which converts a long-term evaluation strategic problem into a number of stage summaries. Not only can the development context be grasped macroscopically, but also the key points with greater changes in the development process can be found. For example, in time, because the situation has changed, the evaluation system and the target of attention have changed greatly. After changing the index system node, the evaluation result changes greatly. After passing the turning point, the strategic situation enters a new stage.

[0057] The device of the application adopts automatic score calculation instead of expert scoring. When the model index system is constructed, key words and groups are added to the leaf node index. The frequency of the appearance of these key words and groups in a large amount of information materials and the number of hit materials are counted. Based on the weighted tf-df (term frequency-document frequency), the index score is converted into a normalized index score. The expert experience judgment is converted into semantic mining based on a large amount of information. The higher the weighted tf-df value, the higher the heat of the index node in the historical materials, the more influential, and the greater the contribution to the total score. For other special indexes, the semantic connotation can be mapped to the number of specific high-value strategic targets. The named entity recognition technology is used to automatically extract the state or number of special objects, and the score is automatically calculated after matching the judgment rules.

[0058] The device of the present application introduces index types and node weights, and adopts automatic score calculation to replace expert scoring. In order to distinguish the contribution of different indexes to the total score, the indexes are divided into "general index-keyword", "general index-rule matching" and "trigger index". Each type of index automatically completes the calculation by mining clues from a large amount of information materials. When constructing the strategic problem index system tree, the key word group, mode matching rule and trigger basic score information of the three types of indexes are respectively input. The "general index-keyword" index counts the frequency of its key word group appearing in the information materials and the number of hit materials, and converts it into a normalized index score based on weighted tf-df (term frequency-document frequency). The expert experience judgment is converted into semantic mining based on massive information. The higher the weighted tf-df value is, the higher the heat of the index node in the historical materials is, the more influential it is, and the greater its contribution to the total score is. The "general index-rule matching" calculates the score according to the matching degree of the extracted structured data and the scoring rule, and the scoring rule is clear and definite without human subjective influence. The trigger index refers to the leaf node that has a strong influence on the total score. Once the situation described by the index occurs, the strategic problem situation will immediately upgrade or decline, and the corresponding total score will directly change. For example, the "XXXX agreement" index item in the "XX situation" strategic problem. Once this situation occurs, the quantitative result will be lower than a certain threshold. The trigger index sets a basic value, the value range of which is -100 to 100. If the marked index occurs, the basic value directly participates in the calculation. Finally, according to the weight of each node in the index system tree, the score of the root node is automatically obtained by recursively weighting and summing the scores of the leaf nodes.

[0059] The device of the present application adopts a periodic evaluation method to track the long-term development trend of the strategic problem. The index system, index key word group and node weight value corresponding to each period can be modified to adapt to the changes of the objects of attention, different focuses and changes of the connotation of the strategic problem in different periods.

[0060] The units described in the embodiments of the present application can be implemented in the form of software or hardware, and the described units can also be arranged in a processor. In some cases, the names of these units do not constitute a limitation on the units themselves.

[0061] According to an aspect of an embodiment of the present application, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to enable the computer device to perform the methods provided in the various optional implementation manners.

[0062] As another aspect, the embodiments of the present application also provide a computer readable medium, which can be included in the electronic device described in the above embodiments, or can exist independently without being assembled into the electronic device. The computer readable medium carries one or more programs, which, when executed by the electronic device, enable the electronic device to implement the method described in the above embodiments.

[0063] The parts of the present application not involved in the embodiments are the same as or can be implemented by the prior art.

[0064] The technical solution described above is only one embodiment of the present application. Based on the application method and principle disclosed in the present application, those skilled in the art can easily make various types of improvements or modifications without being limited to the method described in the above specific embodiments. Therefore, the above description is only preferred and not limiting.

[0065] In addition to the above examples, those skilled in the art can obtain other embodiments by making changes based on the above disclosure or using knowledge or technology in related fields. The features of each embodiment can be interchanged or replaced. Changes and variations made by those skilled in the art do not deviate from the spirit and scope of the present application, and should be within the protection scope of the claims of the present application.

Claims

1. A computer natural language processing-based strategic problem quantification evaluation device, characterized by, The model building unit, the leaf node score calculation unit, the weight setting unit and the evaluation unit are included. The model building unit creates a model of a strategic problem to be analyzed, constructs a strategic problem index system tree, and records index type, index keyword phrase and index matching rule information for each leaf node of the index system tree. The leaf node score calculation unit adopts different leaf node score calculation methods according to the index type. The weight setting unit sets the weight of each node in the index system tree, and obtains the score of the root node, i.e., the quantitative evaluation value of the strategic problem, by layer-by-layer recursive weighted summation based on the leaf node score. The evaluation unit introduces a periodic evaluation method, adjusts the index system, modifies the keyword phrase and matching rule of the leaf node, and recalculates the score according to the actual development of the problem in different periods, to complete the monitoring of the long-term change trend of the strategic problem. The index type includes "general index-keyword", "general index-rule matching" and "trigger index". The different leaf node score calculation methods according to the index type specifically include: The "general index-keyword" is calculated according to the "word frequency-document frequency" of the corresponding keyword phrase in the relevant material text. The "general index-rule matching" is calculated according to the matching degree of the extracted structured data and the scoring rule. The "trigger index" is calculated according to whether it intervenes in the overall quantitative result. In the process of calculating the "general index-rule matching", the rule of each leaf node is freely set; the "general index-rule matching" inputs the scoring formula and substitutes the structured data of the fixed variable to automatically calculate the score. 2.The computer natural language processing based strategic problem quantification evaluation apparatus according to claim 1, wherein, In the process of calculating the "general index-keyword", the "general index-keyword" inputs the keyword phrase, which is a description of the semantic connotation of the index node, and introduces a logical operator between the keywords as the basis for automatic scoring. 3.The computer natural language processing based strategic problem quantification evaluation apparatus according to claim 1, wherein, The "trigger index" is configured with a state bit and a basic value. Once the state bit flag is set to occur, the basic value will directly participate in the calculation of the total score of the root node.

4. The computer natural language processing based strategic problem quantification evaluation apparatus according to claim 1, wherein, The weight setting unit sets the weight value for each node, and the value range is 0 to 1, to distinguish the different contribution degrees of different index items in the total score calculation.

5. The computer natural language processing based strategic problem quantification evaluation apparatus according to claim 1, wherein, The evaluation unit introduces a periodic evaluation method, which converts a long-term strategic problem into numerous stage summaries, can macroscopically grasp the development context, and can also find the key points with large change amplitude in the development process.

6. The strategic problem quantification and evaluation apparatus based on computer natural language processing according to any one of claims 1 to 5, characterized by, The processor and the memory are also included, and the computer program is stored in the memory. When the computer program is loaded and run by the processor, the module architecture of the model building unit, the leaf node score calculation unit, the weight setting unit and the evaluation unit is completed, to monitor the long-term change trend of the strategic problem.

7. The computer natural language processing based strategic problem quantification evaluation apparatus according to claim 6, wherein, The display is also included, and the monitoring of the long-term change trend of the strategic problem is presented on the display.

Images