Big data strategy analysis system

Abstract

The invention relates to a big data strategy analysis system, which solves the technical problems of poor accuracy and low efficiency, and adopts an enterprise portrait analysis unit and a strategy push unit. The edge of the enterprise internal data acquisition and processing unit is arranged in an enterprise, the edge of the enterprise external data acquisition unit is arranged in the enterprise, and the enterprise portrait analysis and processing unit and the portrait result output unit are arranged in the cloud center; the enterprise external data acquisition unit also acquires strategy data; and the enterprise internal data acquisition and processing unit is used for processing the enterprise internal data into parameter score data required by the cloud center enterprise portrait and then transmitting the parameter score data to the enterprise portrait analysis and processing unit, so that the problem is well solved, and the method can be used for accurate strategy pushing.

Description

Technical Field

[0001] The present invention relates to the field of strategy push, and in particular to a big data strategy analysis system. Background Art

[0002] Precision push refers to the new media placing the news, videos, jokes, products, etc. that people want to see in a priority position based on the information processing of big data. Precision push also exists in traditional media, such as different pages of newspapers, different TV channels, different radio bands and different themes, etc.

[0003] Most search engine users enter the same keyword to query, and get the same query results, which cannot distinguish the different needs of different users. The present invention builds an interest model of the enterprise based on the basic data characteristics of the enterprise, the data analysis results, and the search behavior. The platform can also find user groups with similar interests based on the information (strategy, consulting, experts, etc.) that matches the model with its interest model, and realize mutual recommendation. Ultimately, the one-way passive push in the traditional information service model is transformed into a two-way interactive information service, achieving personalized information precision service. Summary of the invention

[0004] The technical problem to be solved by the present invention is the technical problem of poor accuracy and low efficiency in the prior art. A new big data strategy analysis system is provided, which has the characteristics of high accuracy and high efficiency.

[0005] In order to solve the above technical problems, the technical solutions adopted are as follows:

[0006] A big data strategy analysis system, comprising: an enterprise portrait analysis unit and a strategy push unit;

[0007] The edge is set up with an internal data collection and processing unit within the enterprise, an external data collection unit, an enterprise portrait analysis and processing unit set up in the cloud center, and a portrait result output unit; the external data collection unit also collects policy data;

[0008] The enterprise internal data collection and processing unit is used to process the enterprise internal data into parameter score data required for the cloud center enterprise portrait and then transmit it to the enterprise portrait analysis and processing unit;

[0009] The policy push unit is used to execute the policy push program, including the following steps:

[0010] Step 1: Based on the results of the enterprise portrait analysis, enterprises whose similarity with the target enterprise is greater than a predefined threshold are defined as similar enterprises;

[0011] Step 2: Combining the latest updated policies in the policy library after the similar enterprise is recently added to the library, the policies received by the similar enterprise, and the policies used by the similar enterprise as the initial policy push policy;

[0012] Step three: Conditionally match the initial strategy push strategy with the enterprise portrait analysis results, eliminate the strategies in the initial strategy push strategy that do not match the enterprise portrait analysis results, and obtain the precise strategy push strategy.

[0013] The working principle of the present invention is as follows: the interest model of an enterprise is constructed based on the basic data characteristics of the enterprise, the results of data analysis, and the search behavior. The platform can also recommend information (strategies, consultations, experts, etc.) that matches the model with its interest model, and can also find user groups with similar interests to achieve mutual recommendation. Ultimately, the one-way passive push in the traditional information service model is transformed into a two-way interactive information service, achieving personalized information precision service.

[0014] In the above scheme, for optimization, further, the processing of the internal data of the enterprise by the internal data collection and processing unit of the enterprise and the processing of the external data of the enterprise by the cloud center include:

[0015] Step 1: preprocess the data, clean the multi-source heterogeneous data, and convert them into text data;

[0016] Step 2, post-processing the text data to obtain the data ontology, the post-processing includes sentence processing and attribute extraction;

[0017] Step 3: extract the data ontology label, store it as a data knowledge graph, and use the database for storage. The database storage is based on two data types: nodes and relationships. There are several relationships between two nodes in the database, and a relationship has only one start node and an end node.

[0018] Step 4: merge the nodes representing the same entity in several data ontologies and the relationships between the entities to form a fusion graph based on the data ontologies, namely, an enterprise feature graph or a strategy feature graph, including:

[0019] Step 42, calculating the relationship between the final similarity coefficient SIM(e1, e2) and the predefined thresholds t1 and t2. If the final similarity coefficient SIM(e1, e2) ≥ t1, it is defined that the fusion coefficient of entity e1 and entity e2 is greater than the threshold t; if t1 < SIM(e1, e2) < t2, it is defined that the fusion coefficient is equal to the threshold t; if SIM(e1, e2) ≤ t2, it is defined that the fusion coefficient is less than t;

[0020] Step 43, if the fusion coefficient is greater than the threshold t, then fuse entity e1 and entity e2; if it is equal to the threshold t, execute step 44; if it is less than the threshold t, update entity e2 and return to execute step 41;

[0021] Step 44, calculate the structural similarity coefficient SNN(e1, e2) between entity e1 and entity e2, if SNN(e1, e2)>μ*min(Hot(e1), Hot(e2)), define the final similarity SIM(e1, e2)≥t1, return to step 42, otherwise execute step 45, min means taking the smaller value;

[0022]

[0023] Among them, NB(θ1) is the set of neighbor nodes of entity e1, and NB(θ2) is the set of neighbor nodes of entity e2; co(e ′ ,e1) is the node e in the fusion graph ′ The number of co-occurrences with e1, co(e ′ ,e2) is the node e in the fusion graph ′ The number of co-occurrences with e2; e ′ and e1 contribution times, and e ′ The smaller value between the co-occurrence times of e1 and e2 is regarded as entities e1 and e2 share e ′ frequency;

[0024] Step 45, traverse and update entity e2, and return to execute step 41.

[0025] Further, in step 411, entity e1 is defined as character combination s1, entity e2 is required to be character combination s2, and the character combination similarity SFC of entity e1 and entity e2 is calculated. st (e1, e2), such as character combination similarity SFC st (e1, e2)≥predefined threshold t, then define the final similarity SSD(e1, e2)≥t1, otherwise execute step 412;

[0026]

[0027] Where {op1} is the minimum number of steps required to modify the character combination to the required addition, deletion, or modification of characters, len(s1) is the character length of the character combination s1, and len(s2) is the character length of the character combination s2;

[0028] Step 412, define entity e1=w 11 w 12 ... 1m , entity e2 = w 21 w 22 ... 2m , where m>1, n>1 is the number of elements that make up the entity; calculate the semantic similarity SYS se(e1, e2), calculate the final similarity SSD(e1, e2) = SYS se (e1, e2), m and n are two positive integers greater than 1;

[0029]

[0030] Among them, β, α, τ are predefined parameter values; max is the maximum value; Sin is a synonym judgment, Sin(w 1i ) is the element w 1i The set of synonym sets to which it belongs, Sin(w 2j ) is the element w 2j The set of synonym sets to which it belongs, c 1i ,c 2j There are two nodes.

[0031] Furthermore, the enterprise portrait analysis processing unit performs the following steps:

[0032] Step h1, receiving the enterprise internal data indicator score value transmitted by the enterprise internal data collection and processing unit;

[0033] Step h2, receiving the external enterprise data transmitted by the external enterprise data acquisition unit;

[0034] Step h3, calculating the enterprise external data indicator score value based on the enterprise external data;

[0035] Step h4, calculating the difference between the enterprise internal data indicator score and the enterprise external data indicator score, taking the indicators with a difference below the threshold as the first-class indicators, and taking the indicators with a difference above the threshold as the second-class indicators;

[0036] Step h5, optimizing a category of indicators by calculating the average value to obtain a category of final indicator scores;

[0037] Step h6: for the second category of indicators, according to the pre-defined data source weights, a weighted calculation method is used to reconstruct and optimize the final indicator scores of the second category;

[0038] Step h7, input the first category final indicator score and the second category final indicator score into the preset enterprise portrait analysis model to complete the enterprise portrait analysis.

[0039] Further, step h6 comprises:

[0040] Step h61, transmitting the external enterprise data corresponding to the two types of indicators back to the internal data collection and processing unit of the enterprise;

[0041] Step h62, the enterprise internal data collection and processing unit performs verification weighted calculation on the enterprise external data and the enterprise internal data to obtain data reconstruction values ​​of the two types of indicators;

[0042] Step h63, the enterprise internal data collection and processing unit calculates the second category final indicator score of the second category indicator based on the data reconstruction value of the second category indicator.

[0043] Furthermore, similar enterprises identified in step 1 include:

[0044] Step s1: Convert the feature map in the enterprise database into the target feature sequence {X k |k=1,2,3...N}, define the combination of enterprise feature graph sequences in the enterprise database as the reference feature sequence X0;

[0045] Step s2, defined as the length of the sliding caliper, extracts a feature graph sequence from the target feature sequence, where N is the number of feature nodes in the target feature sequence;

[0046] Step s3, using a sliding caliper, in the target feature sequence {X k |k=1,2,3...N}, determine the matching starting feature node, that is, start from the time point when the first feature node sequence of the target feature sequence is received, step 1 feature node to search until N-M+1 feature nodes; set the similarity confidence μ m Exceeding the threshold value f fz The first characteristic node of is taken as the starting sliding position of the sliding caliper, f fz =0.6-0.7;

[0047]

[0048] f(k) is the similarity confidence of a single feature node in the target feature sequence relative to the i-th feature node in the reference feature sequence;

[0049]

[0050] Among them, M is the number of feature nodes in the sliding caliper, that is, the time interval subwindow, and K is the number of feature nodes in the sliding caliper. is the weight of the pth feature parameter, is the target feature sequence {X k The pth characteristic parameter value of the kth characteristic node of |k=1,2,3...N}, where P is the number of characteristic parameters of the characteristic node;

[0051] Step s4, define the tolerance range Δt, the time interval between the feature node and the previous feature node is [t j -Δt,t j+Δt], j = 1, 2, 3..., M; for the target feature sequence, the starting feature node is taken as the time zero point, and the time intervals between the two feature nodes before and after are searched one by one, and the total number of nodes L falling in each time interval sub-window is counted;

[0052] Step s5: If the number of feature nodes satisfies L0≤L≤M, a single feature node multi-parameter matching is performed. If the similarity confidence μ of each corresponding feature node is m Exceeding the threshold value f fz , then the sliding caliper is defined to be matched successfully, otherwise it is defined to be unmatched; where L0≥(M / 2);

[0053] Step s6: If all feature nodes in the reference feature sequence are matched by feature nodes in the sliding caliper of the sequence to be identified, the starting position of the sliding caliper is moved back to the time point when the feature node matched at the rightmost end of the previous window arrives, and the process returns to step s4 with this as the starting point, and the number of matching sliding calipers is increased by 1; if the number of consecutive matching sliding calipers is greater than the threshold N et , the enterprise where the sliding caliper is located is defined as a similar enterprise to the enterprise to be pushed; if the number of matching sliding calipers is insufficient or the multi-parameter matching of a single feature node is unsuccessful, the starting point of the sliding caliper is moved back to the next node position, that is, returning to execute step s3.

[0054] Since the sliding caliper is set, the recognition process does not need to repeatedly traverse each pulse in the entire sequence. It only needs to perform matching comparisons within the sliding caliper and the relative time subwindow. Once the number of matching sliding windows meets the alarm requirements, the recognition can be declared successful, which greatly reduces the recognition time of similar enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0056] Figure 1 ,Schematic diagram of big data strategy analysis system. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0058] Example 1

[0059] This embodiment provides a big data strategy analysis system. Figure 1 , including: enterprise portrait analysis unit and strategy push unit;

[0060] The edge is set up with an internal data collection and processing unit within the enterprise, an external data collection unit, an enterprise portrait analysis and processing unit set up in the cloud center, and a portrait result output unit; the external data collection unit also collects policy data;

[0061] The enterprise internal data collection and processing unit is used to process the enterprise internal data into parameter score data required for the cloud center enterprise portrait and then transmit it to the enterprise portrait analysis and processing unit;

[0062] The policy push unit is used to execute the policy push program, including the following steps:

[0063] Step 1: Based on the results of the enterprise portrait analysis, enterprises whose similarity with the target enterprise is greater than a predefined threshold are defined as similar enterprises;

[0064] Step 2: Combining the latest updated policies in the policy library after the similar enterprise is recently added to the library, the policies received by the similar enterprise, and the policies used by the similar enterprise as the initial policy push policy;

[0065] Step three: Conditionally match the initial strategy push strategy with the enterprise portrait analysis results, eliminate the strategies in the initial strategy push strategy that do not match the enterprise portrait analysis results, and obtain the precise strategy push strategy.

[0066] This embodiment builds an interest model of an enterprise based on the basic data characteristics of the enterprise, the results of data analysis, and the search behavior. The platform can also recommend information (strategies, consulting, experts, etc.) that matches the model with its interest model, and can also find user groups with similar interests to achieve mutual recommendation. Ultimately, the one-way passive push in the traditional information service model is transformed into a two-way interactive information service, achieving personalized information precision service.

[0067] Preferably, the processing of the enterprise internal data by the enterprise internal data collection and processing unit and the processing of the enterprise external data by the cloud center include:

[0068] Step 1: preprocess the data, clean the multi-source heterogeneous data, and convert them into text data;

[0069] Step 2, post-processing the text data to obtain the data ontology, the post-processing includes sentence processing and attribute extraction;

[0070] Step 3: extract the data ontology label, store it as a data knowledge graph, and use the database for storage. The database storage is based on two data types: nodes and relationships. There are several relationships between two nodes in the database, and a relationship has only one start node and an end node.

[0071] Step 4: merge the nodes representing the same entity in several data ontologies and the relationships between the entities to form a fusion graph based on the data ontologies, namely, an enterprise feature graph or a strategy feature graph, including:

[0072] Step 42, calculating the relationship between the final similarity coefficient SIM(e1, e2) and the predefined thresholds t1 and t2. If the final similarity coefficient SIM(e1, e2) ≥ t1, it is defined that the fusion coefficient of entity e1 and entity e2 is greater than the threshold t; if t1 < SIM(e1, e2) < t2, it is defined that the fusion coefficient is equal to the threshold t; if SIM(e1, e2) ≤ t2, it is defined that the fusion coefficient is less than t;

[0073] Step 43, if the fusion coefficient is greater than the threshold t, then fuse entity e1 and entity e2; if it is equal to the threshold t, execute step 44; if it is less than the threshold t, update entity e2 and return to execute step 41;

[0074] Step 44, calculate the structural similarity coefficient SNN(e1, e2) between entity e1 and entity e2, if SNN(e1, e2)>μμmin(Hot(e1), Hot(e2)), define the final similarity SIM(e1, e2)≥t1, return to step 42, otherwise execute step 45, min means taking the smaller value;

[0075]

[0076] Among them, NB(θ1) is the set of neighbor nodes of entity e1, and NB(θ2) is the set of neighbor nodes of entity e2; co(e ′ ,e1) is the node e in the fusion graph ′ The number of co-occurrences with e1, co(e ′ ,e2) is the node e in the fusion graph ′ The number of co-occurrences with e2; e ′ and e1 contribution times, and e ′ The smaller value between the co-occurrence times of e1 and e2 is regarded as entities e1 and e2 share e ′ frequency;

[0077] Step 45, traverse and update entity e2, and return to execute step 41.

[0078] Specifically, step 41 includes:

[0079] Step 411, define entity e1 as character combination s1, require entity e2 as character combination s2, calculate the character combination similarity SFC of entity e1 and entity e2 st (e1, e2), such as character combination similarity SFC st(e1, e2)≥predefined threshold t, then define the final similarity SSD(e1, e2)≥t1, otherwise execute step 412;

[0080]

[0081] Where {op1} is the minimum number of steps required to modify the character combination to the required addition, deletion, or modification of characters, len(s1) is the character length of the character combination s1, and len(s2) is the character length of the character combination s2;

[0082] Step 412, define entity e1=w 11 w 12 ... 1m , entity e2 = w 21 w 22 ... 2m , where m>1, n>1 is the number of elements that make up the entity; calculate the semantic similarity SYS se (e1, e2), calculate the final similarity SSD(e1, e2) = SYS se (e1, e2), m and n are two positive integers greater than 1;

[0083]

[0084] Among them, β, α, τ are predefined parameter values; max is the maximum value; Sin is a synonym judgment, Sin(w 1i ) is the element w 1i The set of synonym sets to which it belongs, Sin(w 2j ) is the element w 2j The set of synonym sets to which it belongs, c 1i ,c 2j There are two nodes.

[0085] Preferably, the enterprise portrait analysis processing unit performs the following steps:

[0086] Step h1, receiving the enterprise internal data indicator score value transmitted by the enterprise internal data collection and processing unit;

[0087] Step h2, receiving the external enterprise data transmitted by the external enterprise data acquisition unit;

[0088] Step h3, calculating the enterprise external data indicator score value based on the enterprise external data;

[0089] Step h4, calculating the difference between the enterprise internal data indicator score and the enterprise external data indicator score, taking the indicators with a difference below the threshold as the first-class indicators, and taking the indicators with a difference above the threshold as the second-class indicators;

[0090] Step h5, optimizing a category of indicators by calculating the average value to obtain a category of final indicator scores;

[0091] Step h6: for the second category of indicators, according to the pre-defined data source weights, a weighted calculation method is used to reconstruct and optimize the final indicator scores of the second category;

[0092] Step h7, input the first category final indicator score and the second category final indicator score into the preset enterprise portrait analysis model to complete the enterprise portrait analysis.

[0093] Preferably, step h6 comprises:

[0094] Step h61, transmitting the external enterprise data corresponding to the two types of indicators back to the internal data collection and processing unit of the enterprise;

[0095] Step h62, the enterprise internal data collection and processing unit performs verification weighted calculation on the enterprise external data and the enterprise internal data to obtain data reconstruction values ​​of the two types of indicators;

[0096] Step h63, the enterprise internal data collection and processing unit calculates the second category final indicator score of the second category indicator based on the data reconstruction value of the second category indicator.

[0097] Preferably, the similar enterprises determined in step 1 include:

[0098] Step s1: Convert the feature map in the enterprise database into the target feature sequence {X k |k=1,2,3...N}, define the combination of enterprise feature graph sequences in the enterprise database as the reference feature sequence X0;

[0099] Step s2, defined as the length of the sliding caliper, extracts a feature graph sequence from the target feature sequence, where N is the number of feature nodes in the target feature sequence;

[0100] Step s3, using a sliding caliper, in the target feature sequence {X k |k=1,2,3...N}, determine the matching starting feature node, that is, start from the time point when the first feature node sequence of the target feature sequence is received, step 1 feature node to search until N-M+1 feature nodes; set the similarity confidence μ m Exceeding the threshold value f fz The first characteristic node of is taken as the starting sliding position of the sliding caliper, f fz =0.6-0.7;

[0101]

[0102] f(k) is the similarity confidence of a single feature node in the target feature sequence relative to the i-th feature node in the reference feature sequence;

[0103]

[0104] Among them, M is the number of feature nodes in the sliding caliper, that is, the time interval subwindow, and K is the number of feature nodes in the sliding caliper. is the weight of the pth feature parameter, is the target feature sequence {X k The pth characteristic parameter value of the kth characteristic node of |k=1,2,3...N}, where P is the number of characteristic parameters of the characteristic node;

[0105] Step s4, define the tolerance range Δt, the time interval between the feature node and the previous feature node is [t j -Δt,t j +Δt], j = 1, 2, 3..., M; for the target feature sequence, the starting feature node is taken as the time zero point, and the time intervals between the two feature nodes before and after are searched one by one, and the total number of nodes L falling in each time interval sub-window is counted;

[0106] Step s5: If the number of feature nodes satisfies L0≤L≤M, a single feature node multi-parameter matching is performed. If the similarity confidence μ of each corresponding feature node is m Exceeding the threshold value f fz , then the sliding caliper is defined to be matched successfully, otherwise it is defined to be unmatched; where L0≥(M / 2);

[0107] Step s6: If all feature nodes in the reference feature sequence are matched by feature nodes in the sliding caliper of the sequence to be identified, the starting position of the sliding caliper is moved back to the time point when the feature node matched at the rightmost end of the previous window arrives, and the process returns to step s4 with this as the starting point, and the number of matching sliding calipers is increased by 1; if the number of consecutive matching sliding calipers is greater than the threshold N et , the enterprise where the sliding caliper is located is defined as a similar enterprise to the enterprise to be pushed; if the number of matching sliding calipers is insufficient or the multi-parameter matching of a single feature node is unsuccessful, the starting point of the sliding caliper is moved back to the next node position, that is, returning to execute step s3.

[0108] In this embodiment, since a sliding caliper is set, the recognition process does not need to repeatedly traverse each pulse in the entire sequence. It only needs to perform matching comparisons within the sliding caliper and the relative time subwindow. Once the number of matching sliding windows meets the alarm requirement, the recognition can be declared successful, which greatly reduces the recognition time of similar enterprises.

[0109] Although the above describes the illustrative specific embodiments of the present invention so that those skilled in the art can understand the present invention, the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, all inventions and creations using the concepts of the present invention are protected.

Claims

1. A big data strategy analysis system, characterized by: The big data strategy analysis system includes: an enterprise portrait analysis unit and a strategy push unit; The edge is set up with an internal data collection and processing unit within the enterprise, an external data collection unit, an enterprise portrait analysis and processing unit set up in the cloud center, and a portrait result output unit; the external data collection unit also collects policy data; The enterprise internal data collection and processing unit is used to process the enterprise internal data into parameter score data required for the cloud center enterprise portrait and then transmit it to the enterprise portrait analysis and processing unit; The policy push unit is used to execute the policy push program, which includes the following steps: Step 1: Based on the results of the enterprise portrait analysis, enterprises whose similarity with the target enterprise is greater than a predefined threshold are defined as similar enterprises; Step 2: Combining the latest updated policies in the policy library after the similar enterprise is recently added to the library, the policies received by the similar enterprise, and the policies used by the similar enterprise as the initial policy push policy; Step three: Conditionally match the initial strategy push strategy with the enterprise portrait analysis results, eliminate the strategies in the initial strategy push strategy that do not match the enterprise portrait analysis results, and obtain the precise strategy push strategy.

2. The big data strategy analysis system according to claim 1, characterized in that: The processing of internal data by the enterprise data collection and processing unit and the processing of external data by the cloud center include: Step 1: preprocess the data, clean the multi-source heterogeneous data, and convert them into text data; Step 2, post-processing the text data to obtain the data ontology, the post-processing includes sentence processing and attribute extraction; Step 3: extract the data ontology label, store it as a data knowledge graph, and use the database for storage. The database storage is based on two data types: nodes and relationships. There are several relationships between two nodes in the database, and a relationship has only one start node and an end node. Step 4: merge the nodes representing the same entity in several data ontologies and the relationships between the entities to form a fusion graph based on the data ontologies, namely, an enterprise feature graph or a strategy feature graph, including: Step 41, using a similarity determination method based on ontology elements, select entity e1 and entity e2, and calculate the final similarity coefficient SIM(e1, e2) between the two; Step 42, calculating the relationship between the final similarity coefficient SIM(e1, e2) and the predefined thresholds t1 and t2. If the final similarity coefficient SIM(e1, e2) ≥ t1, it is defined that the fusion coefficient of entity e1 and entity e2 is greater than the threshold t; if t1 < SIM(e1, e2) < t2, it is defined that the fusion coefficient is equal to the threshold t; if SIM(e1, e2) ≤ t2, it is defined that the fusion coefficient is less than t; Step 43, if the fusion coefficient is greater than the threshold t, then fuse entity e1 and entity e2; if it is equal to the threshold t, execute step 44; if it is less than the threshold t, update entity e2 and return to execute step 41; Step 44, calculate the structural similarity coefficient SNN(e1, e2) between entity e1 and entity e2, if SNN(e1, e2)>μ*min(Hot(e1), Hot(e2)), define the final similarity SIM(e1, e2)≥t1, return to step 42, otherwise execute step 45, min means taking the smaller value; Among them, NB(θ1) is the set of neighbor nodes of entity e1, NB(θ2) is the set of neighbor nodes of entity e2; co(e ′ ,e1) is the node e in the fusion graph ′ The number of co-occurrences with e1, co(e ′ ,e2) is the node e in the fusion graph ′ The number of co-occurrences with e2; e ′ and e1 contribution times, and e ′ The smaller value between the co-occurrence times of e1 and e2 is regarded as entities e1 and e2 share e ′ frequency; Step 45, traverse and update entity e2, and return to execute step 41.

3. The big data strategy analysis system according to claim 2, characterized in that: Step 41 includes: Step 411, define entity e1 as character combination s1, require entity e2 as character combination s2, calculate the character combination similarity SFC of entity e1 and entity e2 st (e1, e2), such as character combination similarity SFC st (e1, e2)≥predefined threshold t, then define the final similarity SSD(e1, e2)≥t1, otherwise execute step 412; Where {op1} is the minimum number of steps required to modify the character combination to the required addition, deletion, or modification of characters, len(s1) is the character length of the character combination s1, and len(s2) is the character length of the character combination s2; Step 412, define entity e1=w 11 w 12 ... 1m , entity e2 = w 21 w 22 ... 2m , where m>1, n>1 is the number of elements that make up the entity; calculate the semantic similarity SYS se (e1, e2), calculate the final similarity SSD(e1, e2) = SYS se (e1, e2), m and n are two positive integers greater than 1; Among them, β, α, τ are predefined parameter values; max is the maximum value; Sin is a synonym judgment, Sin(w 1i ) is the element w 1i The set of synonym sets to which it belongs, Sin(w 2j ) is the element w 2j The set of synonym sets to which it belongs, c 1i ,c 2j There are two nodes.

4. The big data strategy analysis system according to claim 1, characterized in that: The enterprise portrait analysis processing unit performs the following: Step h1, receiving the enterprise internal data indicator score value transmitted by the enterprise internal data collection and processing unit; Step h2, receiving the external enterprise data transmitted by the external enterprise data acquisition unit; Step h3, calculating the enterprise external data indicator score value based on the enterprise external data; Step h4, calculating the difference between the enterprise internal data indicator score and the enterprise external data indicator score, taking the indicators with a difference below the threshold as the first-class indicators, and taking the indicators with a difference above the threshold as the second-class indicators; Step h5, optimizing a category of indicators by calculating the average value to obtain a category of final indicator scores; Step h6: for the second category of indicators, according to the pre-defined data source weights, a weighted calculation method is used to reconstruct and optimize the final indicator scores of the second category; Step h7, input the first category final indicator score and the second category final indicator score into the preset enterprise portrait analysis model to complete the enterprise portrait analysis.

5. The big data strategy analysis system according to claim 4, characterized in that: Step h6 includes: Step h61, transmitting the external enterprise data corresponding to the two types of indicators back to the internal data collection and processing unit of the enterprise; Step h62, the enterprise internal data collection and processing unit performs verification weighted calculation on the enterprise external data and the enterprise internal data to obtain data reconstruction values ​​of the two types of indicators; Step h63, the enterprise internal data collection and processing unit calculates the second category final indicator score of the second category indicator based on the data reconstruction value of the second category indicator.

6. The big data strategy analysis system according to claim 2, characterized in that: Similar enterprises identified in step 1 include: Step s1: Convert the feature map in the enterprise database into the target feature sequence {X k |k=1,2,3...N}, define the combination of enterprise feature graph sequences in the enterprise database as the reference feature sequence X0; Step s2, defined as the length of the sliding caliper, extracts a feature graph sequence from the target feature sequence, where N is the number of feature nodes in the target feature sequence; Step s3, using a sliding caliper, in the target feature sequence {X k |k=1,2,3...N}, determine the matching starting feature node, that is, start from the time point when the first feature node sequence of the target feature sequence is received, step 1 feature node to search until N-M+1 feature nodes; set the similarity confidence μ m Exceeding the threshold value f fz The first characteristic node of is used as the starting sliding position of the sliding caliper; f(k) is the similarity confidence of a single feature node in the target feature sequence relative to the i-th feature node in the reference feature sequence; Among them, M is the number of feature nodes in the sliding caliper, that is, the time interval subwindow, and K is the number of feature nodes in the sliding caliper. is the weight of the pth feature parameter, is the target feature sequence {X k The pth characteristic parameter value of the kth characteristic node of |k=1,2,3...N}, where P is the number of characteristic parameters of the characteristic node; Step s4, define the tolerance range Δt, the time interval between the feature node and the previous feature node is [t j -Δt,t j +Δt], j = 1, 2, 3..., M; for the target feature sequence, the starting feature node is taken as the time zero point, and the time intervals between the two feature nodes before and after are searched one by one, and the total number of nodes L falling in each time interval sub-window is counted; Step s5: If the number of feature nodes satisfies L0≤L≤M, a single feature node multi-parameter matching is performed. If the similarity confidence μ of each corresponding feature node is m Exceeding the threshold value f fz , then the sliding caliper is defined to be matched successfully, otherwise it is defined to be unmatched; where L0≥(M / 2); Step s6: If all feature nodes in the reference feature sequence are matched by feature nodes in the sliding caliper of the sequence to be identified, the starting position of the sliding caliper is moved back to the time point when the feature node matched at the rightmost end of the previous window arrives, and the process returns to step s4 with this as the starting point, and the number of matching sliding calipers is increased by 1; if the number of consecutive matching sliding calipers is greater than the threshold N et , the enterprise where the sliding caliper is located is defined as a similar enterprise to the enterprise to be pushed; if the number of matching sliding calipers is insufficient or the multi-parameter matching of a single feature node is unsuccessful, the starting point of the sliding caliper is moved back to the next node position, that is, returning to execute step s3.

Images