CN106910148B - Collaborative filtering-based adaptive pushing method for command elements - Google Patents

Collaborative filtering-based adaptive pushing method for command elements Download PDF

Info

Publication number
CN106910148B
CN106910148B CN201710044622.8A CN201710044622A CN106910148B CN 106910148 B CN106910148 B CN 106910148B CN 201710044622 A CN201710044622 A CN 201710044622A CN 106910148 B CN106910148 B CN 106910148B
Authority
CN
China
Prior art keywords
command
commander
similarity
battle
command element
Prior art date
Legal status (The legal status 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 status listed.)
Active
Application number
CN201710044622.8A
Other languages
Chinese (zh)
Other versions
CN106910148A (en
Inventor
崔翛龙
杜波
袁琛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Individual
Original Assignee
Individual
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 Individual filed Critical Individual
Priority to CN201710044622.8A priority Critical patent/CN106910148B/en
Publication of CN106910148A publication Critical patent/CN106910148A/en
Application granted granted Critical
Publication of CN106910148B publication Critical patent/CN106910148B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services
    • G06Q50/265—Personal security, identity or safety
    • 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

Abstract

The invention relates to the technical field of personalized recommendation, in particular to a command element self-adaptive pushing method based on collaborative filtering, which comprises the first step of constructing an anti-terrorism resource pool and storing and recording comprehensive real-time updating data of anti-terrorism actions; secondly, constructing pre-collaborative filtering based on the fighting type-command element and sending a calculation result to a commander; and thirdly, constructing two-stage collaborative filtering based on commander-command elements and sending the calculated structure to the commander. The method filters command element data which are possibly required by the anti-terrorist battle tasks of different types, so that a commander can conveniently search related information according to the loaded battle tasks, the problem that the commander faces a large amount of data and has no need to take the next place is solved, and the anti-terrorist data push efficiency is improved; the method has the advantages that the most needed data are respectively pushed according to the roles of the commanders in the anti-terrorism battle, so that the difference of the commanders in different levels on the data requirements in the same anti-terrorism battle is solved, and the accuracy of pushing the anti-terrorism data is improved.

Description

Collaborative filtering-based adaptive pushing method for command elements
Technical Field
The invention relates to the technical field of personalized recommendation, in particular to a command element self-adaptive pushing method based on collaborative filtering.
Background
At present, aiming at the characteristics of diversification of anti-terrorist action tasks, quantization of command elements and networking of command levels, the direct construction of the collaborative filtering relationship between a commander and the command elements has to face the dilemma of data sparseness, cold start and the like, so that the efficiency is low, and the accuracy is difficult to ensure. The anti-terrorism command control platform is a data interaction hub for implementing anti-terrorism actions by each command node, and has great significance for improving the data flow transfer rate and further improving the anti-terrorism action assistant decision-making capability by researching the information interaction mode of the command control platform. At present, with the rapid increase of global data volume, the problem of information overload is more and more serious, and a great amount of redundant information causes that a commander cannot acquire effective command element information in time, thereby influencing the action decision efficiency.
Collaborative filtering is one of representative technologies of recommendation systems, and the application field is very wide. The basic principle is that the interest of a target user is predicted according to the interest of a neighbor user, and then corresponding commodities are recommended to the target user. From the classification point of view, the currently common collaborative filtering recommendation techniques mainly include: collaborative filtering recommendations based on users, collaborative filtering recommendations based on items, collaborative filtering recommendations based on models, collaborative filtering recommendations based on temporal weighting, and hybrid collaborative filtering recommendations, among others. From the perspective of recommendation effect, collaborative filtering recommendation generally has the problems of data sparsity, cold start, instantaneity and the like, and directly influences the efficiency and accuracy of data recommendation, so that many domestic and foreign students develop research on the problems and provide some improvement and optimization strategies. The methods are improved in the aspects of similarity judgment, initial data reduction processing, score improvement method and the like, so that the recommendation precision is improved. The anti-terrorism command control system is used as a special recommendation system and has the characteristics of various data types, complex information marshalling, high real-time requirement and the like. Therefore, aiming at the characteristics of multiple combat missions, multiple information elements, multiple instruction levels and the like of anti-terrorism actions, the development of the collaborative filtering algorithm research has certain guiding significance on the engineering practice of the development of the instruction control system.
Disclosure of Invention
The invention provides a collaborative filtering-based command element self-adaptive pushing method, overcomes the defects of the prior art, and can effectively solve the problems of data sparseness and cold start in the traditional recommendation algorithm.
The technical scheme of the invention is realized by the following measures: the command element self-adaptive pushing method based on collaborative filtering comprises the following steps:
firstly, constructing an anti-terrorism resource pool, and storing and recording comprehensive real-time updating data of anti-terrorism actions;
secondly, constructing pre-collaborative filtering based on the fighting type-command element and sending a calculation result to a commander;
thirdly, constructing a scoring matrix based on two-stage collaborative filtering of the commander-command elements, comprising the following processes:
(1) a two-dimensional m multiplied by n scoring matrix is constructed by adopting commander-command elements, and the structure of the two-dimensional m multiplied by n scoring matrix is as follows:
Figure GDA0002602129450000011
wherein T represents the battle mission, I represents the command element, m represents the number of the battle mission T, and n represents the number of the command element I;
(2) the score R is obtained by weighted calculation according to the historical application probability of the command elements in the battle mission and the importance score of the anti-terrorist expertsijThe calculation method is shown in the following formula:
Figure GDA0002602129450000021
xijfor battle missions TiFor command element IjHistorical click times, ∑ xijFor battle missions TiTotal number of times of (c), xij/∑xijFor battle missions TiApplication command element IjProbability of yijFor fighting task T by anti-terrorist expertsiApplication command element IjThe importance score of (1), the score yij∈(0,1)；1、2Are respectively command elements IjIn battle mission TiThe historical application probability and the weight of the anti-terrorist expert importance score in the step (1); sigmakIs the sum of the weights, and k is a natural number;
the fourth step, calculate the similarity between the commanders based on the aggregation subset, includes the following steps:
(1) determining similarity among the commanders and defining related constraint conditions, wherein the specific definition is as follows:
a. defining the similarity relationship between the commanders:
Csim＝κHie(u,v)+Area(u,v)+ξTime(u,v) (2)
wherein u and v respectively represent commanders u and v, variables kappa and xi are weight coefficients, and the weight coefficients satisfy 0 ≦ kappa, xi, < 1 and kappa + xi + ═ 1;
b. the constraints on the command level, command territory and battle time are defined as follows:
the command level Hie (u, v) is set to 7 levels, which respectively correspond to a headquarter, a branch team, a big team, a middle team, a team and a class, and command relation formulas of different command levels are as follows:
Figure GDA0002602129450000022
wherein (0.1,0.2,0.3, 0.4,0.6,0.8, 1) represents the weight of the 7-level membership;
secondly, the command region Area (u, v) represents the fighting region relation of the commanders u and v, and the formula is as follows:
Figure GDA0002602129450000023
wherein 1 represents being in the same combat zone, 0 represents being in a different combat zone, AreauIndicates the battle Area, where the commander u is locatedvRepresenting the combat zone in which commander v is located;
③ the fighting Time (u, v) represents the fighting Time relationship of the requests sent by the commanders u, v, and the formula is as follows:
Figure GDA0002602129450000024
wherein, 1 represents that the two pieces of equipment are in the same fighting time interval when sending out the request, and 0 represents that the two pieces of equipment are in different fighting time intervals when sending out the request; timeuIndicating the request period, Time, of the commander uvRepresents a request period of commander v;
(2) constructing a similarity matrix UU of the commander according to the formula (2)AThe similarity relation transmission between the commanders is measured by adopting similarity matrix multiplication, namely UUA×UUARepresenting the relationship of two-step reachable indirect similarity, and marking the two-step reachable indirect similarity matrix after matrix operation as UU ', wherein each element is UU'ijRepresenting the two-step reachable similarity of the commander i and the commander j, the similarity between the commanders is the sum of the direct similarity and the two-step reachable similarity, marking as UUA, and carrying out descalement and dimensionization on the similarity matrix, namely
Figure GDA0002602129450000031
Wherein:
Figure GDA0002602129450000032
UUA′ijthe similarity after standard deviation transformation is represented, the mean value is 0, and the standard deviation is 1;
(3) to ensure UUA'ijIn the interval [0,1]And (4) carrying out standard deviation transformation by internal fluctuation, wherein the formula is as follows:
Figure GDA0002602129450000033
wherein, URijRepresenting the final similarity of the commander i and the commander j, and UR is a final similarity matrix;
fifthly, a calculation formula based on the improved modified cosine similarity is as follows:
Figure GDA0002602129450000034
Figure GDA0002602129450000035
wherein R isi,cAnd Rj,cRespectively represents the scores, x, of the commander i and the commander j on the command element information Ci,cThe historical request times, y, of the command element typei,cThe value is scored for the expert and,1、2is a weighted value; URijIndicating fingerFinal similarity of the commander i and the commander j;
Figure GDA0002602129450000036
representing the average value of the scores of the commander i and the commander j for the command element information; u shapeijRepresenting a set of generic command elements; u shapeiRepresenting a commander i command element set; u shapejRepresenting a set of commander j command elements; sim (u)i,uj) The modified cosine similarity of the commander i and the commander j is integrally understood;
sixthly, calculating two-stage recommendation filtering prediction scores, giving a threshold value beta, and determining the commanders as the neighbors of the target commander when the similarity between the commanders is greater than the threshold value beta, wherein the prediction scoring method formula is as follows:
Figure GDA0002602129450000037
and (3) deducing beta e (0,1) according to a similarity calculation formula (8), giving scores of the commander for the unused command element information according to a formula (10), pushing the M command element information with the front scores to the target commander, and ending.
The following is further optimization or/and improvement of the technical scheme of the invention:
in the second step, the pre-collaborative filtering based on the battle type-command element comprises the following steps:
(1) the third step is the same as the method for constructing the battle type-command element information scoring matrix;
(2) calculating and correcting cosine similarity, expressing the scores of all command elements by the battle mission by using an n-dimensional vector, expressing the importance score of the battle mission on the ith command element by using the ith component in the vector, and setting the score vector of a target battle mission u as
Figure GDA0002602129450000038
The score vector of the battle mission v is
Figure GDA0002602129450000039
The cosine similarity between the two battle types is as follows:
Figure GDA0002602129450000041
wherein R isuiScoring the importance of the target combat task u to the command element i, RviScoring the importance of the battle mission v to the command element i, wherein i belongs to (1, n), the similarity between the battle missions u and v is equal to the cosine value of the included angle between the scoring vectors, and the value range is [0, 1%](ii) a K combat missions with the maximum similarity are selected to form a nearest neighbor set of the target combat mission; for a target combat task u, a set of nearest neighbors Nu＝{v1,v2,……,vKTherein of
Figure GDA0002602129450000046
viAccording to similarity sim (u, v)i) (i is more than or equal to 1 and less than or equal to K) are arranged in sequence from large to small;
(3) the 'battle type-command element' recommends prediction calculation, and a nearest neighbor set N of a target battle type u is found according to the similarity of the battle types in pre-collaborative filteringuThen, for the target battle type u, the importance scores of the target command element i by the nearest neighbor v are weighted and summed to obtain the importance prediction score P of u to iuiThe calculation formula is as follows:
Figure GDA0002602129450000042
wherein sim (u, v) is the battle type similarity calculated by battle types u and v according to the cosine similarity measurement method used in the previous stage, RviScoring the command elements i for similar combat missions v;
(4) generating filtering recommendation, and recommending the command elements with the first N prediction scores to a target combat type to form a new target combat type vector:
Figure GDA0002602129450000043
wherein the content of the first and second substances,
Figure GDA0002602129450000044
the vector is a target battle type vector subjected to first collaborative filtering, M is the number of command elements used by the target battle type, N is the number of command elements recommended by filtering, the sum of the number of command elements and the number of command elements recommended by filtering is the number y of basic command elements subjected to next filtering, and any I in the vector is a command element subjected to one-time filtering.
The method also comprises a seventh step of analyzing and calculating the error, wherein the calculation formula is as follows:
Figure GDA0002602129450000045
wherein i represents the ith scoring result, N is a natural number, and N represents the scoring number; the set of scores obtained by prediction is denoted as { p }1,p2,…,pnThe actual evaluation set is q1,q2,…,qn}; and generating data according to the grading result, sending the data to the anti-terrorism resource pool, correcting the pre-collaborative filtering and the two-stage collaborative filtering, and ending.
According to the invention, through a multiple collaborative filtering algorithm MCFM facing an anti-terrorist task, the algorithm firstly excavates the collaborative relationship between the fighting type and the command information element through pre-collaborative filtering P-CFA; then, a social network analysis SNA condensation subgroup analysis method is integrated into a traditional collaborative filtering algorithm for improving a similarity calculation method, so that the problems of data sparseness and cold start are effectively solved, the accurate and real-time pushing of anti-terrorism action command element information is realized, and the anti-terrorism fighting capacity based on an information system is improved. The multiple collaborative filtering algorithm can better adapt to the pushing requirement of command element information in the command control system, and can improve the autonomous recommendation efficiency and recommendation accuracy of the system to a certain extent. The method filters command element data which are possibly required by the anti-terrorist battle tasks of different types, so that a commander can conveniently search related information according to the loaded battle tasks, the problem that the commander faces a large amount of data and has no need to take the next place is solved, and the anti-terrorist data push efficiency is improved; the method has the advantages that the most needed data are respectively pushed according to the roles of the commanders in the anti-terrorism battle, so that the difference of the commanders in different levels on the data requirements in the same anti-terrorism battle is solved, and the accuracy of pushing the anti-terrorism data is improved.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
FIG. 2 is a schematic representation of the similarity relationship propagation of the present invention.
Fig. 3.1 is a schematic diagram of a second-order reachable relationship network of the target commander captain according to embodiment 2 of the present invention.
Fig. 3.2 is a schematic diagram of a second-order reachability relationship network of a captain in the target commander in embodiment 2 of the present invention.
Fig. 3.3 is a schematic diagram of a second-order reachable relationship network for the target commander growth elimination in embodiment 2 of the present invention.
FIG. 4 is a diagram showing the influence of the parameter β in example 2 of the present invention.
FIG. 5 is a schematic comparison of the MCFM and IBCF algorithms of example 2 of the present invention.
FIG. 6 is a graph showing the comparison of the MCFM and IBCF algorithm efficiency of example 2 of the present invention.
Detailed Description
The present invention is not limited by the following examples, and specific embodiments may be determined according to the technical solutions and practical situations of the present invention.
The invention is further described with reference to the following examples and figures:
example 1: as shown in fig. 1 and 2, the collaborative filtering-based command element adaptive pushing method includes the following steps:
firstly, constructing an anti-terrorism resource pool, and storing and recording comprehensive real-time updating data of anti-terrorism actions;
secondly, constructing pre-collaborative filtering based on the fighting type-command element and sending a calculation result to a commander;
thirdly, constructing a scoring matrix based on two-stage collaborative filtering of the commander-command elements, comprising the following processes:
(1) a two-dimensional m multiplied by n scoring matrix is constructed by adopting commander-command elements, and the structure of the two-dimensional m multiplied by n scoring matrix is as follows:
Figure GDA0002602129450000051
wherein T represents the battle mission, I represents the command element, m represents the number of the battle mission T, and n represents the number of the command element I;
(2) the score R is obtained by weighted calculation according to the historical application probability of the command elements in the battle mission and the importance score of the anti-terrorist expertsijThe calculation method is shown in the following formula:
Figure GDA0002602129450000052
xijfor battle missions TiFor command element IjHistorical click times, ∑ xijFor battle missions TiTotal number of times of (c), xij/∑xijFor battle missions TiApplication command element IjProbability of yijFor fighting task T by anti-terrorist expertsiApplication command element IjThe importance score of (1), the score yij∈(0,1)；1、2Are respectively command elements IjIn battle mission TiThe historical application probability and the weight of the anti-terrorist expert importance score in the step (1);
herein, the1、2According to the current situation and rule of the anti-terrorism battle in the current stage, the sum sigma is obtained by solving by an AHP methodkIs the sum of the weights. Battle mission T in scoring matrixiFor command element IjScore R ofijE (0,1), the higher score represents the fighting task TiChinese command element IjThe higher the degree of importance.
The fourth step, calculate the similarity between the commanders based on the aggregation subset, includes the following steps:
(2) determining similarity among the commanders and defining related constraint conditions, wherein the specific definition is as follows:
a. defining similarity relationships between commanders
Csim＝κHie(u,v)+Area(u,v)+ξTime(u,v) (2)
Wherein u and v respectively represent commanders u and v, variables kappa and xi are weight coefficients, and the weight coefficients satisfy 0 ≦ kappa, xi, < 1 and kappa + xi + ═ 1; the variables k, xi respectively influence the proportion of the command level Hie (u, v), the command Area (u, v) and the fighting Time (u, v) in similarity calculation, and the invention takes k as xi as 1/3;
b. the constraints on the command level, command territory and battle time are defined as follows:
the command level Hie (u, v) is set to 7 levels, which respectively correspond to a headquarter, a branch team, a big team, a middle team, a team and a class, and command relation formulas of different command levels are as follows:
Figure GDA0002602129450000061
wherein (0.1,0.2,0.3, 0.4,0.6,0.8, 1) represents the weight of the 7-level membership;
secondly, the command region Area (u, v) represents the fighting region relation of the commanders u and v, and the formula is as follows:
Figure GDA0002602129450000062
wherein 1 represents being in the same combat zone, 0 represents being in a different combat zone, AreauIndicates the battle Area, where the commander u is locatedvRepresenting the combat zone in which commander v is located;
③ the fighting Time (u, v) represents the fighting Time relationship of the requests sent by the commanders u, v, and the formula is as follows:
Figure GDA0002602129450000063
wherein, 1 represents that the requests are in the same fighting time interval, and 0 represents that the requests are in different fighting time intervals; timeuIndicating the request period, Time, of the commander uvRepresents a request period of commander v;
(2) constructing a similarity matrix UU of the commander according to the formula (2)AThe similarity relation transmission between the commanders is measured by adopting similarity matrix multiplication, namely UUA×UUARepresenting the relationship of two-step reachable indirect similarity, and marking the two-step reachable indirect similarity matrix after matrix operation as UU ', wherein each element is UU'ijRepresenting the two-step reachable similarity of the commander i and the commander j, the similarity between the commanders is the sum of the direct similarity and the two-step reachable similarity, marking as UUA, and carrying out descalement and dimensionization on the similarity matrix, namely
Figure GDA0002602129450000064
Wherein:
Figure GDA0002602129450000065
UUA′ijthe similarity after standard deviation transformation is represented, the mean value is 0, and the standard deviation is 1;
(3) to ensure UUA'ijIn the interval [0,1]And (4) carrying out standard deviation transformation by internal fluctuation, wherein the formula is as follows:
Figure GDA0002602129450000066
wherein, URijRepresenting the final similarity of the commander i and the commander j, and UR is a final similarity matrix;
through the formula (2), a similarity matrix UU of the commander can be constructedA. In fig. 2, reachability among nodes is generally expressed by matrix multiplication, and is known by a director through finding a similarity relationship, and similarity between a target director and other directors is determined by multiple reachable paths, that is, a similarity transfer relationship exists, including direct similarity and direct similarityIndirect similarity, i.e. hierarchical command and override command relation in anti-terrorist action, as shown in figure 2, in similarity calculation based on coagulation subgroup analysis, similarity relation transfer between commanders can be measured by similarity matrix multiplication, i.e. UUA×UUAThe two-step reachable indirect similarity relation is represented, and based on the SNA 2-party indirect relation research theory, the two-step reachable indirect similarity relation between the commanders is only considered in the invention, and the two-step reachable indirect similarity here represents the override command relation. Marking the two-step reachable indirect similarity matrix after matrix operation as UU ', wherein each element of UU'ijRepresenting two-step reachable similarity of commander i to commander j. Therefore, the similarity between the commanders is the sum of the direct similarity and the two-step reachable similarity, and is marked as UUA.
A Social Network Analysis (SNA) method is a structured Analysis method that mines the internal structure of a Social Network by extracting relationship patterns in Social members. Coagulation subgroup analysis is an important research direction for social network analysis. The main body of the anti-terrorist action is a commander, the command relationship among different commanders is clear, the command flow is clear, and an obvious hierarchical structure is presented, and the structural trend of flattening and networking of the command relationship becomes obvious with the increasing diversity of combat types. The invention classifies the commander by utilizing the coacervate subgroup with larger theoretical value and application value, and aims to construct the coacervate subgroup based on the user similarity relationship.
Fifthly, a calculation formula based on the improved modified cosine similarity is as follows:
Figure GDA0002602129450000071
Figure GDA0002602129450000072
wherein R isi,cAnd Rj,cRespectively represents the scores, x, of the commander i and the commander j on the command element information Ci,cThe historical request times, y, of the command element typei,cThe value is scored for the expert and,1、2for weighting value, the invention is easy to calculate1＝2＝1/2；URijRepresenting the final similarity of commander i to commander j;
Figure GDA0002602129450000073
representing the average value of the scores of the commander i and the commander j for the command element information;
here the similarity calculation is crucial in the collaborative filtering algorithm, which determines the accuracy of the recommendation. Typically characterized by cosine similarity, Pearson correlation coefficients, and modified cosine similarity. Considering that the consistency of the cosine similarity to the scoring scale is considered to have defects, and the Pearson correlation coefficient needs to consider factors such as the correlation coefficient among samples, the modified cosine similarity is adopted for calculation in the pre-collaborative filtering.
Sixthly, calculating two-stage recommendation filtering prediction scores, giving a threshold value beta, and determining the commanders as the neighbors of the target commander when the similarity between the commanders is greater than the threshold value beta, wherein the prediction scoring method formula is as follows:
Figure GDA0002602129450000074
and (3) deducing beta e (0,1) according to a similarity calculation formula (8), giving scores of the commander for the unused command element information according to a formula (10), and further pushing M command element information with the scores higher to the commander. Here, the formula (8) is based on the calculation of the modified cosine similarity, and thus the value range of the threshold β can be determined.
According to the invention, through a multiple collaborative filtering algorithm MCFM facing an anti-terrorist task, the algorithm firstly excavates the collaborative relationship between the fighting type and the command information element through pre-collaborative filtering P-CFA; then, a social network analysis SNA condensation subgroup analysis method is integrated into a traditional collaborative filtering algorithm for improving a similarity calculation method, so that the problems of data sparseness and cold start are effectively solved, accurate and real-time pushing of anti-terrorist action command element information is realized, and the anti-terrorist fighting capacity based on an information system is improved. The multiple collaborative filtering algorithm can better adapt to the pushing requirement of command element information in the command control system, and can improve the autonomous recommendation efficiency and recommendation accuracy of the system to a certain extent. The method filters command element data which are possibly required by the anti-terrorist battle tasks of different types, so that a commander can conveniently search related information according to the loaded battle tasks, the problem that the commander faces a large amount of data and has no need to take the next place is solved, and the anti-terrorist data push efficiency is improved; the method has the advantages that the most needed data are respectively pushed according to the roles of the commanders in the anti-terrorism battle, so that the difference of the commanders in different levels on the data requirements in the same anti-terrorism battle is solved, and the accuracy of pushing the anti-terrorism data is improved.
The command element self-adaptive pushing method based on collaborative filtering can be further optimized or/and improved according to actual needs:
as shown in fig. 1, in the second step, the pre-collaborative filtering based on the battle mission-command elements comprises the following steps:
(1) the third step is the same as the method for constructing the battle type-command element information scoring matrix;
(2) calculating and correcting cosine similarity, expressing the scores of all command elements by the battle mission by using an n-dimensional vector, expressing the importance score of the battle mission on the ith command element by using the ith component in the vector, and setting the score vector of a target battle mission u as
Figure GDA0002602129450000081
The score vector of the battle mission v is
Figure GDA0002602129450000082
The cosine similarity between the two battle types is as follows:
Figure GDA0002602129450000083
wherein R isuiCommand element i for target battle task u pairImportance score of RviScoring the importance of the battle mission v to the command element i, wherein i belongs to (0, 1-n), the similarity between the battle missions u and v is equal to the cosine value of the included angle between the scoring vectors, and the value range is
[0,1]. And selecting the K battle missions with the maximum similarity to form a nearest neighbor set of the target battle missions. For a target combat task u, a set of nearest neighbors Nu＝{v1,v2,……,vKTherein of
Figure GDA0002602129450000085
viAccording to similarity sim (u, v)i) (i is more than or equal to 1 and less than or equal to K) are arranged in sequence from large to small;
through similarity calculation, a battle mission similarity matrix P of n multiplied by n orders is obtained, the matrix is a symmetric matrix, and each element in the matrix represents the similarity degree between two corresponding battle missions. Since each task does not calculate similarity with itself, the elements on the diagonal of the matrix are all 0, and the similarity values of each row in P are sorted in descending order.
And calculating the correlation between the battle tasks according to the scoring matrix of the fighting types and the command factors. In the scoring matrix, each row represents a scoring vector for each combat mission, where cosine similarity is used as a measure of similarity between the target combat mission and other combat missions. The cosine similarity describes the degree of correlation between two vectors by calculating the included angle between the two vectors, the smaller the included angle is, the more similar the two vectors are represented, and the larger the included angle is, the more dissimilar the two vectors are represented.
(3) The 'battle type-command element' recommends prediction calculation, and a nearest neighbor set N of a target battle type u is found according to the similarity of the battle types in cooperative filteringuThen, for the target battle type u, the importance scores of the target command element i by the nearest neighbor v are weighted and summed to obtain the importance prediction score P of u to iuiThe calculation formula is as follows:
Figure GDA0002602129450000084
wherein sim (u, v) is the battle type similarity calculated by battle types u and v according to the cosine similarity measurement method used in the previous stage, RviScoring the command elements i for similar combat missions v;
(4) generating filtering recommendation, and recommending the command elements with the first N prediction scores to a target combat type to form a new target combat type vector:
Figure GDA0002602129450000091
wherein the content of the first and second substances,
Figure GDA0002602129450000092
the vector is a target battle type vector subjected to first collaborative filtering, M is the number of command elements used by the target battle type, N is the number of command elements recommended by filtering, the sum of the number of command elements and the number of command elements recommended by filtering is the number y of basic command elements subjected to next filtering, and any I in the vector is a command element subjected to one-time filtering.
And a seventh step of analyzing and calculating errors according to requirements, wherein the calculation formula is as follows:
Figure GDA0002602129450000093
wherein the set of scores obtained by prediction is denoted as { p }1,p2,…,pnThe actual evaluation set is q1,q2,…,qn}; and generating data according to the grading result, sending the data to the anti-terrorism resource pool, correcting the pre-collaborative filtering and the two-stage collaborative filtering, and ending.
Example 2: as shown in fig. 3.1, 3.2, 3.3, 4, 5 and 6, two terrorists hijack 3 people in the city of W by holding guns, drive a minibus and try to collide with ZS road to temporarily check roadblocks, which causes paralysis of partial road sections in the city area and extreme panic of citizens. The terrorists flee to the bridge of the textile mill after being blocked by traffic police, abandon the vehicle and enter the residential buildings nearby, and attempt to confront with police by virtue of the landform advantages, so that the situation is very critical. The joint-fingered command armed police W city squad quickly enters a task area and completes the tasks of sealing, controlling, deploying, field disposing and hostage rescuing; the teams report according to the situation, and plan that 20 officers and soldiers are dispatched by the 1 and 3 middle teams to handle the task, and the assistant teams can be used for carrying out the overall command. After the task region is reached, the subsidiary captain C, the captain E in the queue 1 and the captain F in the queue 3 respectively request task related data from the command center through the command terminal so as to further draw up a detailed operation plan; the command element self-adaptive pushing method based on collaborative filtering in armed police army performing tasks comprises the following steps:
the method comprises the steps of firstly, extracting data from an anti-terrorism data resource pool, wherein the data comprise 15 anti-terrorism combat styles of hijacking hostage, individual attack, riot (harassment) disorder and the like, 120 command elements of terrorist quantity, equipment, social situations, weather, terrain and the like, requests of 2300 command elements are 12000 times, and scores of 20000 records given by experts to the command elements according to the request times and related relations are distributed in an interval of [0,5], and the higher the score is, the stronger the demand of the commander to a specific command element under a certain combat style is. The data are stored according to headquarter A, headquarter B, branch C, big D, middle E, rank F, class G and 7 command levels in a grading way. In the embodiment of the invention, data such as establishing a collaborative filtering matrix, determining a correlation relationship and the like are calculated on the basis of the data, and the correlation data are subjected to decryption processing;
secondly, experimental design and result analysis, wherein similarity relation networks between command levels constructed based on the condensation subset analysis are shown in the attached figures 3.1, 3.2 and 3.3;
the Agent1-Agent7 respectively represent headquarter A, headquarter B, branch queue C, big queue D, middle queue E, row F and shift G7 command levels, all levels are connected by bidirectional arrows to represent that the information flow is interactive, the similarity index Csim is calculated by a formula (2), the second-order reachable relation of the similarity relation is obtained by matrix multiplication, and finally the dimensionless processing is carried out according to formulas (6) and (7) to obtain the similarity index Csim;
third, parameter beta-shadowChanging a similarity judgment parameter beta in two-stage collaborative filtering in a sound test, and taking N1＝3，N2Respectively taking 10, 20 and 30, and calculating results through simulation as shown in the figure 4; the following conclusions can be drawn from fig. 4:
(a) number of neighbors N2Next, as the beta value increases, the change trend of the recommendation error MAE is substantially consistent;
(b) beta is 0.3 as an optimal threshold value, when the beta is less than 0.3, the error is reduced along with the increase of the beta value, and when the beta is more than 0.3, the error is positively correlated with the beta;
(c) error and N at the same value of beta2Negative correlation when N2Above 20 the amplification decreases. For this reason, we can choose beta less than 0.3, N in practical application 220, thereby ensuring that the highest recommendation precision and recommendation efficiency are achieved under the condition of meeting the combat requirement; the optimal value of the parameter beta is found by observing the influence of the parameter beta on the algorithm of the invention.
Fourthly, comparing the recommended accuracy of MCFM and IBCF
On the basis of the third test (taking the best test parameters as the parameters involved), a comparative test was carried out with a conventional project-Based collaborative filtering algorithm (Item Based-CF) as a control group, and the results are shown in FIG. 5;
as can be seen from fig. 5, compared with the conventional project-based collaborative filtering algorithm, the task-oriented multiple collaborative filtering algorithm provided by the present invention has smaller recommendation errors MAE under different command information element numbers, so that the algorithm provided by the present invention effectively improves the recommendation precision and can improve the information transmission and push quality of the command control system;
fifthly, comparing MCFM and IBCF recommended efficiency
Selecting different command element numbers, and performing a comparative test by taking a traditional project-Based collaborative filtering algorithm (Item Based-CF) as a control group, wherein the result is shown in figure 6;
as can be seen from fig. 6, as the number of the "neighbor projects", that is, the command information, increases, the average running time of the conventional project-based collaborative filtering algorithm increases sharply, while the running time of the task-oriented multiple collaborative filtering algorithm of the present invention is relatively stable and much lower than that of the conventional algorithm; the multiple collaborative filtering algorithm provided by the invention is efficient and stable, can adapt to retrieval, filtering and recommendation under data of different scales, and has an obvious effect on improving the recommendation efficiency of the command control system.
And sixthly, recommending a result to a commander, wherein the algorithm recommending result is as follows:
Isubsidiary captainRoad traffic, terrain, local crowd's reaction to events }
I1 team leader-available resources }for terrain, road traffic
I3 middle queue length-terrain, available resources, road traffic }
Verification shows that the multiple collaborative filtering algorithm provided by the invention can be used for commanding relevant data of command elements required by individual recommendation of anti-terrorism combat for commanders of different levels.
The technical characteristics form an embodiment of the invention, which has strong adaptability and implementation effect, and unnecessary technical characteristics can be increased or decreased according to actual needs to meet the requirements of different situations.

Claims (3)

1. A command element self-adaptive pushing method based on collaborative filtering is characterized by comprising the following steps:
firstly, constructing an anti-terrorism resource pool, and storing and recording comprehensive real-time updating data of anti-terrorism actions;
secondly, constructing pre-collaborative filtering based on the fighting type-command element and sending a calculation result to a commander;
thirdly, constructing a scoring matrix based on two-stage collaborative filtering of the commander-command elements, comprising the following processes:
(1) a two-dimensional m multiplied by n scoring matrix is constructed by adopting commander-command elements, and the structure of the two-dimensional m multiplied by n scoring matrix is as follows:
Figure FDA0002602129440000011
wherein T represents the battle mission, I represents the command element, m represents the number of the battle mission T, and n represents the number of the command element I;
(2) the score R is obtained by weighted calculation according to the historical application probability of the command elements in the battle mission and the importance score of the anti-terrorist expertsijThe calculation method is shown in the following formula:
Figure FDA0002602129440000012
xijfor battle missions TiFor command element IjHistorical click times, ∑ xijFor battle missions TiTotal number of times of (c), xij/∑xijFor battle missions TiApplication command element IjProbability of yijFor fighting task T by anti-terrorist expertsiApplication command element IjThe importance score of (1), the score yij∈(0,1)；1、2Are respectively command elements IjIn battle mission TiThe historical application probability and the weight of the anti-terrorist expert importance score in the step (1); sigmak
Is the sum of the weights, and k is a natural number;
the fourth step, calculate the similarity between the commanders based on the aggregation subset, includes the following steps:
(1) determining similarity among the commanders and defining related constraint conditions, wherein the specific definition is as follows:
a. defining the similarity relationship between the commanders:
Csim＝κHie(u,v)+Area(u,v)+ξTime(u,v) (2)
wherein u and v respectively represent commanders u and v, variables kappa and xi are weight coefficients, and the weight coefficients satisfy 0 ≦ kappa, xi, < 1 and kappa + xi + ═ 1;
b. the constraints on the command level, command territory and battle time are defined as follows:
the command level Hie (u, v) is set to 7 levels, which respectively correspond to a headquarter, a branch team, a big team, a middle team, a team and a class, and command relation formulas of different command levels are as follows:
Figure FDA0002602129440000013
wherein (0.1,0.2,0.3, 0.4,0.6,0.8, 1) represents the weight of the 7-level membership;
secondly, the command region Area (u, v) represents the fighting region relation of the commanders u and v, and the formula is as follows:
Figure FDA0002602129440000021
wherein 1 represents being in the same combat zone, 0 represents being in a different combat zone, AreauIndicates the battle Area, where the commander u is locatedvRepresenting the combat zone in which commander v is located;
③ the fighting Time (u, v) represents the fighting Time relationship of the requests sent by the commanders u, v, and the formula is as follows:
Figure FDA0002602129440000022
wherein, 1 represents that the two pieces of equipment are in the same fighting time interval when sending out the request, and 0 represents that the two pieces of equipment are in different fighting time intervals when sending out the request; timeuIndicating the request period, Time, of the commander uvRepresents a request period of commander v;
(2) constructing a similarity matrix UU of the commander according to the formula (2)AThe similarity relation transmission between the commanders is measured by adopting similarity matrix multiplication, namely UUA×UUARepresenting the relationship of two-step reachable indirect similarity, and marking the two-step reachable indirect similarity matrix after matrix operation as UU ', wherein each element is UU'ijRepresenting the two-step reachable similarity of the commander i and the commander j, the similarity between the commanders is the sum of the direct similarity and the two-step reachable similarity, marking as UUA, and carrying out descalement and dimensionization on the similarity matrix, namely
Figure FDA0002602129440000023
Wherein:
Figure FDA0002602129440000024
UUA′ijthe similarity after standard deviation transformation is represented, the mean value is 0, and the standard deviation is 1;
(3) to ensure UUA'ijIn the interval [0,1]And (4) carrying out standard deviation transformation by internal fluctuation, wherein the formula is as follows:
Figure FDA0002602129440000025
wherein, URijRepresenting the final similarity of the commander i and the commander j, and UR is a final similarity matrix;
fifthly, a calculation formula based on the improved modified cosine similarity is as follows:
Figure FDA0002602129440000026
Figure FDA0002602129440000027
wherein R isi,cAnd Rj,cRespectively represents the scores, x, of the commander i and the commander j on the command element information Ci,cThe historical request times, y, of the command element typei,cThe value is scored for the expert and,1、2is a weighted value; URijRepresenting the final similarity of commander i to commander j;
Figure FDA0002602129440000028
Figure FDA0002602129440000031
to represent commandersThe average value of the command element information of the i and the commander j is scored; u shapeijRepresenting a set of generic command elements; u shapeiRepresenting a commander i command element set; u shapejRepresenting a set of commander j command elements; sim (u)i,uj) The modified cosine similarity of the commander i and the commander j is integrally understood;
sixthly, calculating two-stage recommendation filtering prediction scores, giving a threshold value beta, and determining the commanders as the neighbors of the target commander when the similarity between the commanders is greater than the threshold value beta, wherein the prediction scoring method formula is as follows:
Figure FDA0002602129440000032
and (3) deducing beta e (0,1) according to a similarity calculation formula (8), giving scores of the commander for the unused command element information according to a formula (10), pushing the M command element information with the front scores to the target commander, and ending.
2. The collaborative filtering-based command element adaptive pushing method according to claim 1, wherein in the second step, the pre-collaborative filtering based on the battle type-command element comprises the following steps:
(1) the third step is the same as the method for constructing the battle type-command element information scoring matrix;
(2) calculating and correcting cosine similarity, expressing the scores of all command elements by the battle mission by using an n-dimensional vector, expressing the importance score of the battle mission on the ith command element by using the ith component in the vector, and setting the score vector of a target battle mission u as
Figure FDA0002602129440000033
The score vector of the battle mission v is
Figure FDA0002602129440000034
The cosine similarity between the two battle types is as follows:
Figure FDA0002602129440000035
wherein R isuiScoring the importance of the target combat task u to the command element i, RviScoring the importance of the battle mission v to the command element i, wherein i belongs to (1, n), the similarity between the battle missions u and v is equal to the cosine value of the included angle between the scoring vectors, and the value range is [0, 1%](ii) a K combat missions with the maximum similarity are selected to form a nearest neighbor set of the target combat mission; for a target combat task u, a set of nearest neighbors Nu＝{v1,v2,……,vKTherein of
Figure FDA0002602129440000039
viAccording to similarity sim (u, v)i) (i is more than or equal to 1 and less than or equal to K) are arranged in sequence from large to small;
(3) the 'battle type-command element' recommends prediction calculation, and a nearest neighbor set N of a target battle type u is found according to the similarity of the battle types in pre-collaborative filteringuThen, for the target battle type u, the importance scores of the target command element i by the nearest neighbor v are weighted and summed to obtain the importance prediction score P of u to iuiThe calculation formula is as follows:
Figure FDA0002602129440000036
wherein sim (u, v) is the battle type similarity calculated by battle types u and v according to the cosine similarity measurement method used in the previous stage, RviScoring the command elements i for similar combat missions v;
(4) generating filtering recommendation, and recommending the command elements with the first N prediction scores to a target combat type to form a new target combat type vector:
Figure FDA0002602129440000037
wherein the content of the first and second substances,
Figure FDA0002602129440000038
the vector is a target battle type vector subjected to first collaborative filtering, M is the number of command elements used by the target battle type, N is the number of command elements recommended by filtering, the sum of the number of command elements and the number of command elements recommended by filtering is the number y of basic command elements subjected to next filtering, and any I in the vector is a command element subjected to one-time filtering.
3. The collaborative filtering-based command element adaptive pushing method according to claim 1 or 2, further comprising a seventh step of performing error analysis calculation according to the following calculation formula:
Figure FDA0002602129440000041
wherein i represents the ith scoring result, N is a natural number, and N represents the scoring number; the set of scores obtained by prediction is denoted as { p }1,p2,…,pnThe actual evaluation set is q1,q2,…,qn}; and generating data according to the grading result, sending the data to the anti-terrorism resource pool, correcting the pre-collaborative filtering and the two-stage collaborative filtering, and ending.
CN201710044622.8A 2017-01-19 2017-01-19 Collaborative filtering-based adaptive pushing method for command elements Active CN106910148B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710044622.8A CN106910148B (en) 2017-01-19 2017-01-19 Collaborative filtering-based adaptive pushing method for command elements

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710044622.8A CN106910148B (en) 2017-01-19 2017-01-19 Collaborative filtering-based adaptive pushing method for command elements

Publications (2)

Publication Number Publication Date
CN106910148A CN106910148A (en) 2017-06-30
CN106910148B true CN106910148B (en) 2020-11-17

Family

ID=59206581

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710044622.8A Active CN106910148B (en) 2017-01-19 2017-01-19 Collaborative filtering-based adaptive pushing method for command elements

Country Status (1)

Country Link
CN (1) CN106910148B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110298546A (en) * 2019-05-23 2019-10-01 北京邮电大学 A kind of method for allocating tasks and device

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101685458A (en) * 2008-09-27 2010-03-31 华为技术有限公司 Recommendation method and system based on collaborative filtering
CN102508870A (en) * 2011-10-10 2012-06-20 南京大学 Individualized recommending method in combination of rating data and label data
CN102541920A (en) * 2010-12-24 2012-07-04 华东师范大学 Method and device for improving accuracy degree by collaborative filtering jointly based on user and item
CN103412948A (en) * 2013-08-27 2013-11-27 北京交通大学 Cluster-based collaborative filtering commodity recommendation method and system
CN106326390A (en) * 2016-08-17 2017-01-11 成都德迈安科技有限公司 Recommendation method based on collaborative filtering

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101685458A (en) * 2008-09-27 2010-03-31 华为技术有限公司 Recommendation method and system based on collaborative filtering
CN102541920A (en) * 2010-12-24 2012-07-04 华东师范大学 Method and device for improving accuracy degree by collaborative filtering jointly based on user and item
CN102508870A (en) * 2011-10-10 2012-06-20 南京大学 Individualized recommending method in combination of rating data and label data
CN103412948A (en) * 2013-08-27 2013-11-27 北京交通大学 Cluster-based collaborative filtering commodity recommendation method and system
CN106326390A (en) * 2016-08-17 2017-01-11 成都德迈安科技有限公司 Recommendation method based on collaborative filtering

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
《协同过滤算法中数据稀疏性问题研究》;代金龙;《中国优秀硕士学位论文全文数据库 信息科技辑》;20140315;I138-1110 *

Also Published As

Publication number Publication date
CN106910148A (en) 2017-06-30

Similar Documents

Publication Publication Date Title
CN112329348B (en) Intelligent decision-making method for military countermeasure game under incomplete information condition
Looney et al. Cognitive situation and threat assessments of ground battlespaces
CN110197282B (en) Threat estimation and situation assessment method based on genetic fuzzy logic tree
Gautam et al. TOPSIS for multi criteria decision making in intuitionistic fuzzy environment
CN107944694A (en) A kind of equipment Safeguard operational process analysis method towards operational performance
Radovanović et al. Application of the fuzzy AHP-VIKOR hybrid model in the selection of an unmanned aircraft for the needs of tactical units of the armed forces
Hostetler et al. Inferring strategies from limited reconnaissance in real-time strategy games
CN106910148B (en) Collaborative filtering-based adaptive pushing method for command elements
CN113177583B (en) Aerial target clustering method
Yin et al. A semi-Markov decision model for recognizing the destination of a maneuvering agent in real time strategy games
Wang et al. Air combat effectiveness assessment of military aircraft using a fuzzy AHP and TOPSIS methodology
CN112926739A (en) Network countermeasure effectiveness evaluation method based on neural network model
Barry et al. Simulation in context; using data farming for decision support
CN113673773B (en) Learning path recommendation method integrating knowledge background and learning time prediction
Hicks et al. Intelligent agent-based software architecture for combat performance under overwhelming information inflow and uncertainty
Wang et al. Time series prediction with incomplete dataset based on deep bidirectional echo state network
Wang et al. Learning deep hidden nonlinear dynamics from aggregate data
CN114565027A (en) Threat degree evaluation method based on aerial flight target radar data
Frey et al. Tactical situation modelling of MUM-T helicopter mission scenarios using influence maps
Belue Multilayer perceptrons for classification
Lauren A fractal-based approach to equations of attrition
Poropudas et al. Influence diagrams in analysis of discrete event simulation data
Kashef Forecasting The Price of the Flight Tickets using A Novel Hybrid Learning model
CN115952428B (en) Group task identification method based on GRU
CN117436282B (en) Architecture optimization method based on dual variable weight and TOPSIS-gray correlation

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant