Rank aggregation method and apparatus, device, and storage medium

Abstract

The application discloses a ranking aggregation method and device, equipment and a storage medium, and the method comprises the following steps: acquiring ranking information; constructing a user similarity relationship network based on the ranking information, and decomposing the user similarity relationship network into a plurality of high-density communities according to a preset community detection algorithm; obtaining behavior structure characteristics of the high-density communities based on the ranking information, and determining ranking weights of users in the high-density communities according to the behavior structure characteristics of the high-density communities; and calculating ranking aggregation results of an object based on the ranking weights and the ranking information according to a preset ranking aggregation method. The ranking results of the users on the object are analyzed, the high-density communities among the users are found, the ranking weights of the high-density communities are obtained, the ranking aggregation results are obtained according to the ranking information and the ranking weights of the users in the high-density communities, the ranking weights determined by the behavior structure characteristics of the colluding users are used to reflect the effectiveness of the ranking results of the colluding users, and the authenticity of the ranking aggregation results is improved.

Description

Technical Field

[0001] The present invention relates to the field of ranking aggregation technology, and in particular to a ranking aggregation method and apparatus, device, and storage medium. Background Art

[0002] Ranking is an intuitive and effective way to present choices and evaluations. Ranking problems arise in many fields, such as university or discipline rankings by professional evaluation agencies, ranking scientists in academic fields, and ranking the likelihood of disease-causing genes in biomedical engineering. In most cases, decision makers often need to consider ranking criteria for different attributes or the preferences of different judges to obtain a consistent ranking of objects. Consolidating numerous single ranking lists into a single, consistent ranking that considers all information is also known as the ranking aggregation problem.

[0003] At present, due to the lack of an effective screening mechanism, most ranking aggregation methods have poor robustness against abnormal data such as malicious negative reviews and water army ratings. Especially when a large number of users collude to attack, that is, multiple users publish similar ranking results after prior consultation, traditional ranking aggregation methods will fail, causing these abnormal data to become "mainstream opinions" in the information fusion process, thereby controlling the ranking results. Summary of the Invention

[0004] Based on this, it is necessary to propose a ranking aggregation method and device, equipment and storage medium to address the above problems in order to improve the reliability of the ranking results.

[0005] To achieve the above objectives, the present application provides a ranking aggregation method in a first aspect, the method comprising:

[0006] Obtaining ranking information, wherein the ranking information includes a ranking result of each user for a preset object;

[0007] Constructing a user similarity relationship network based on the ranking information, and decomposing the user similarity relationship network into a plurality of high-density communities according to a preset community detection algorithm;

[0008] Obtaining behavioral structural characteristics of the high-density community based on the ranking information, and determining ranking weights of users in the high-density community according to the behavioral structural characteristics of the high-density community;

[0009] The ranking aggregation result of the object is calculated using the ranking weight and the ranking information based on a preset ranking aggregation method.

[0010] Furthermore, constructing a user similarity relationship network based on the ranking information specifically includes:

[0011] Normalizing the ranking information to obtain ranking data of the object by each user;

[0012] Calculating user similarity between two users based on the ranking data of the objects by each user, wherein the user similarity is the similarity between the ranking results of two users;

[0013] Use users as nodes and similarities as edges to build a user similarity relationship network.

[0014] Furthermore, the calculating of user similarities between two users based on the ranking data of the objects by the respective users specifically includes:

[0015] Obtain a first ranking object set of a first user and a second ranking object set of a second user, wherein the first user and the second user are any one of all users;

[0016] taking the union of the first ranking object set and the second ranking object set as a third ranking object set;

[0017] Calculate, based on the ranking data of the objects by the respective users and the third ranking object set, the ranking similarity between the first user and the second user for each ranking object in the third ranking object set;

[0018] The user similarity between the two users is calculated based on the ranking similarity of the first user and the second user to each ranking object in the third ranking object set.

[0019] Furthermore, the user similarity relationship network is decomposed into a number of high-density communities according to a preset community detection algorithm, specifically including:

[0020] Taking each node in the user similarity relationship network as a community, calculating the modularity gain between the first node and the community where each other node is located for the i-th time, where the first node is any one of all nodes and the initial value of i is 1;

[0021] When there is a positive modularity gain between the first node and the communities where each other node is located, the first node is moved to the community where the second node is located to obtain an aggregated community, wherein the second node is the node corresponding to the maximum modularity gain among the modularity gains between the first node and the communities where each other node is located;

[0022] Each of the aggregated communities is regarded as a new node, with i=i+1, and the step of calculating the modularity gain between the first node and the communities where other nodes are located is continued for the i-th time until a preset stopping condition is met, thereby obtaining several high-density communities in the user similarity relationship network.

[0023] Furthermore, determining the ranking weights of users in the high-density community according to the behavioral structure characteristics of the high-density community specifically includes:

[0024] Calculating a preset collusion index according to the behavioral structural characteristics of the high-density community to obtain collusion index data of the high-density community;

[0025] The ranking weight of each user in the high-density community is calculated based on the collusion index data.

[0026] Furthermore, the step of calculating a preset collusion index based on the behavioral structural characteristics of the high-density community to obtain the collusion index data of the high-density community specifically includes:

[0027] Obtaining a target ranking object set within a target high-density community according to the behavioral structure characteristics of the high-density community, and ranking data of the target ranking objects combined by users within the target high-density community, wherein the target high-density community is any one of all high-density communities, and the target ranking object set is a set of objects ranked by two or more users within the target high-density community;

[0028] Obtaining ranking data of the target ranking object set from other communities except the target high-density community;

[0029] Calculating the deviation of the ranking of the target high-density community based on the ranking data of the target ranking object set in the target high-density community and the ranking data of the target ranking object set in other communities to obtain the ranking deviation of the target high-density community;

[0030] obtaining the number of users who rank a target ranking object in the target high-density community based on the behavioral structure characteristics of the high-density community, where the target ranking object is any one of the target ranking object set;

[0031] Obtaining the total number of users who have ranked the target ranking object;

[0032] Calculate a ratio based on the number of users who rank the target ranking object in the target high-density community and the total number of users to obtain a user ratio of the target high-density community;

[0033] Determining the community size of the target high-density community according to the behavioral structural characteristics of the high-density community;

[0034] The ranking deviation, user ratio and community size of the target high-density community constitute the collusion indicator data of the target high-density community.

[0035] Furthermore, the ranking weight of each user in the high-density community is calculated by the following formula:

[0036]

[0037] Where W(g) is the ranking weight of the target high-density community, RD(g) is the ranking deviation of the target high-density community, UR(g) is the user ratio of the target high-density community, GS(g) is the community size of the target high-density community, and U g is the number of users in the target high-density community.

[0038] To achieve the above-mentioned purpose, the second aspect of the present application provides a ranking aggregation device, the device comprising: an information acquisition unit, a network decomposition unit and the ranking aggregation unit;

[0039] The information acquisition unit is used to acquire ranking information, wherein the ranking information includes the ranking result of each user for the preset object;

[0040] The network decomposition unit is used to construct a user similarity relationship network based on the ranking information, and decompose the user similarity relationship network into a plurality of high-density communities according to a preset community detection algorithm;

[0041] The ranking aggregation unit is configured to obtain the behavioral structure characteristics of the high-density community based on the ranking information, and determine the ranking weights of users in the high-density community according to the behavioral structure characteristics of the high-density community;

[0042] The ranking aggregation result of the object is calculated using the ranking weight and the ranking information based on a preset ranking aggregation method.

[0043] To achieve the above-mentioned purpose, the third aspect of the present application provides a computer-readable storage medium storing a computer program, characterized in that when the computer program is executed by a processor, the processor executes the steps of the method described in the first aspect.

[0044] To achieve the above-mentioned objectives, the fourth aspect of the present application provides a computer device, comprising a memory and a processor, characterized in that the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method described in the first aspect.

[0045] The embodiments of the present invention have the following beneficial effects:

[0046] An embodiment of the present invention proposes a ranking aggregation method, comprising: obtaining ranking information, the ranking information including each user's ranking results for a preset object; constructing a user similarity relationship network based on the ranking information, and decomposing the user similarity relationship network into a number of high-density communities according to a preset community detection algorithm; obtaining the behavioral structural characteristics of the high-density communities based on the ranking information, and determining the ranking weights of the users in the high-density communities based on the behavioral structural characteristics of the high-density communities; and calculating the ranking aggregation results of the objects using the ranking weights and ranking information based on a preset ranking aggregation method. By analyzing the ranking results of users for the objects, high-density communities between users are found, and appropriate ranking weights are determined based on the behavioral structural characteristics of the high-density communities. The ranking information and ranking weights of the users within the high-density communities are then aggregated to obtain a ranking aggregation result. The ranking weights determined based on the behavioral structural characteristics of the colluding users reflect the validity of the collusive ranking results, thereby improving the authenticity of the ranking aggregation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] in:

[0049] Figure 1 Schematic diagram of the flow of the ranking aggregation method in an embodiment of the present invention;

[0050] Figure 2 A user similarity relationship network in an embodiment of the present invention;

[0051] Figure 3 This is a structural block diagram of a ranking aggregation device according to an embodiment of the present invention;

[0052] Figure 4 2 is a diagram showing the internal structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0054] Rankings provide an intuitive display of choices and evaluations, making it easier for users to compare and choose. However, when collecting user ranking information on objects, we discovered that a large number of users may be maliciously posting negative reviews or using online scams to manipulate ratings for specific objects. This phenomenon is referred to as a user collusion attack, and multiple users who publish similar ranking results after prior negotiation are considered suspected collusion attack groups.

[0055] To address the problem of collusion attacks, the present invention discloses a ranking aggregation method that obtains user similarity networks and uses community mining to identify high-density communities in the user similarity networks, i.e., suspected collusion attack groups. For users suspected of collusion attacks, measures such as reducing their weights are taken to eliminate the impact of abnormal data on the ranking aggregation results. For details, please refer to Figure 1 , Figure 1 : This is a flow chart of a ranking aggregation method according to an embodiment of the present invention. The method includes:

[0056] Step 110: Obtain ranking information, which includes each user's ranking result for a preset object.

[0057] In one embodiment of the present invention, assume that there are N users and M objects, and the users can rank some or all of the objects. Let R = (r ui ) N×M Represents the ranking matrix obtained, that is, ranking information. Among them, r ui ≤L u , r ui represents the ranking of object i in the ranking results of user u, L u =|r ui |r ui >0, 1≤i≤M, 1≤u≤N, L u Indicates the ranking length of user u’s ranking result. If user u does not rank object i, then r ui = 0; if user u ranks object i and object j the same, then r ui =r uj .

[0058] Step 120: construct a user similarity relationship network based on the ranking information, and decompose the user similarity relationship network into several high-density communities according to a preset community detection algorithm.

[0059] In an embodiment of the present invention, similarity analysis is performed on the ranking results of objects by each user to determine the similarity of the ranking results among all users, and a user similarity relationship network is constructed with each user as a node and the similarity between two users as an edge.

[0060] High-density communities have highly similar rankings for users and their ranking results for the same users, making them suspected of collusion. Since users in high-density communities are highly likely to engage in malicious behavior when ranking, reducing the ranking weight of users in high-density communities can eliminate the impact of abnormal data on the aggregate ranking results, making the aggregate ranking results more reliable.

[0061] To identify high-density communities within user similarity networks, i.e., suspected groups of collusion and attack, embodiments of the present invention employ a pre-defined community detection algorithm to mine these high-density communities within user similarity networks. This algorithm then determines the weights of the ranking results for users within these high-density communities based on their ranking information, thereby improving the reliability of the ranking results for users within these high-density communities.

[0062] Step 130: Obtain behavioral structural features of the high-density community based on the ranking information, and determine the ranking weights of users in the high-density community according to the behavioral structural features of the high-density community.

[0063] In the embodiment of the present invention, the behavioral characteristics and structural characteristics such as ranking objects, ranking results, and number of users in the high-density community can be obtained based on the ranking information analysis.

[0064] The behavioral structure characteristics of high-density communities can determine whether the ranking behavior of users in high-density communities is abnormal. For example, if users in a high-density community provide similar / identical rankings for a specific object, it indicates that the users are conspiring to attack the object or that water armies are scalping the rankings. If a specific object is mainly ranked by users in a high-density community, and other users outside the community rarely participate in the ranking, it may also indicate that users are conspiring to attack the object or that water armies are scalping the rankings. Most communities consisting of 2 or 3 users may be formed by chance, while large communities consisting of many users are often formed in collusion in advance.

[0065] In one embodiment of the present invention, the malicious behavior degree of the high-density community is determined by the behavioral structure characteristics of the high-density community, and an appropriate ranking weight is determined based on the malicious behavior degree, so as to perform ranking aggregation based on the ranking weight.

[0066] Step 140: Calculate the ranking aggregation result of the object using the ranking weight and ranking information based on a preset ranking aggregation method.

[0067] In the embodiment of the present invention, users belonging to the same high-density community have the same ranking weight. The ranking aggregation method can be Borda Count method, Dowdall method, etc., which are not limited here.

[0068] The ranking aggregation method aggregates the ranking results and ranking weights of multiple users into a comprehensive ranking list, which can be expressed as:

[0069]

[0070] Where w u represents the ranking weight of user u, R u represents the ranking result of user u, Represents the ranking aggregation method, and then the ranking r of object i in the ranking results of user u ui Can be replaced by r ui w u .

[0071] The present invention establishes an overall association between users through their ranking results and obtains a corresponding user similarity network; identifies high-density communities in the user similarity network, i.e., suspected collusion attack groups, through community mining, calculates the ranking weight of each community, and then aggregates the ranking results of multiple users based on the ranking weights, which can effectively eliminate the influence of abnormal data such as malicious reviews and collusion attacks, making the aggregated ranking results more accurate and reliable.

[0072] In one embodiment of the present invention, step 120, constructing a user similarity relationship network based on ranking information, specifically includes:

[0073] Step 121: Normalize the ranking information to obtain the ranking data of each user for the object.

[0074] In this embodiment of the present invention, a user can rank some or all of the objects. Therefore, the ranking results are linearly mapped to the interval [0, 1] using maximum and minimum value normalization to obtain the normalized user ranking data of the objects. The ranking data can be calculated using the following formula:

[0075]

[0076] Where r ui represents the ranking of object i in the ranking results of user u, max(R u ) and min(R u ) represent the ranking results R of user u respectively u The maximum and minimum ranking in .

[0077] Step 122: Calculate the user similarity between two users based on the ranking data of each user on the object. The user similarity is the similarity between the ranking results of two users.

[0078] By analyzing the ranking data of each object by each user, the similarity of the ranking results between any two users can be obtained.

[0079] In one embodiment of the present invention, Step 122, calculating user similarity between two users based on ranking data of each user on the object, specifically includes:

[0080] A1. Obtain a first ranking object set of a first user and a second ranking object set of a second user, wherein the first user and the second user are any one of all users.

[0081] Specifically, the ranking object sets of any two users are obtained. For example, the ranking object set of user u is (A, B, C, D), and the ranking object set of user v is (A, B, C).

[0082] A2. The union of the first ranking object set and the second ranking object set is used as the third ranking object set.

[0083] Specifically, take the union of the ranking object sets of two users. For example, if the ranking object set of user u is (A, B, C, D) and the ranking object set of user v is (A, B, C), then the third ranking object set is (A, B, C).

[0084] A3. Calculate the ranking similarity between the first user and the second user for each ranked object in the third ranked object set based on the ranking data of each user for the object and the third ranked object set.

[0085] In the embodiment of the present invention, the ranking similarity between two users for each third ranking object is determined based on the ranking data of the two users for the third ranking object.

[0086] Specifically, the ranking similarity between two users can be obtained according to the following formula:

[0087]

[0088] Where s(u,v,i) is the ranking similarity between user u and user v on object i, w(r i ″) is the similarity weight, Represents the ranking data of user u for object i, Represents the ranking data of user v for object i.

[0089] In another embodiment of the present invention, a method for determining similarity weights based on ranking data is introduced. The similarity weights can be obtained according to the following formula:

[0090]

[0091] r i ″=min(r u ' i ,r v ' i ),

[0092]

[0093] In the formula, α is a positive constant that determines the weight w(r i ″) with r i The rate of increase and decrease of ″, min(r u ' i ,r v ' i ) represents the ranking r u ' i and r v ' i The minimum value in r u ' i 、r v ' i Indicates how far a high rank is from the top of the ranked list or how far a low rank is from the bottom of the ranked list. This similarity weight reflects that deviations between high-ranked or low-ranked objects are more important than deviations between objects in the middle.

[0094] A4. Calculate the user similarity between the two users based on the ranking similarity of the first user and the second user for each ranking object in the third ranking object set.

[0095] Specifically, after obtaining the ranking similarity of the first user and the second user for each ranking object in the third ranking object set, the user similarity between the two users, that is, the similarity between the ranking results of the two users, can be further calculated.

[0096] In one embodiment of the invention, the user similarity between two users is calculated using the following formula:

[0097]

[0098] Where S(u,v) is the user similarity between user u and user v, I u and I v They represent the ranking object sets of user u and user v, namely the first ranking object set and the second ranking object set respectively.

[0099] Step 123: Use users as nodes and similarities as edges to build a user similarity relationship network.

[0100] Since the ranking results published by colluding users are identical or similar, in order to identify colluding users, an embodiment of the present invention constructs a user similarity relationship network based on user similarity, where users are nodes and similar relationships are edges. User similarity is defined as the weight of the edge, and only edges with weights greater than a preset threshold β are retained.

[0101] In one embodiment of the present invention, in order to identify high-density communities in a user similarity relationship network, a Louvain community detection algorithm may be used. Specifically, step 130 decomposes the user similarity relationship network into a plurality of high-density communities according to a preset community detection algorithm, specifically including:

[0102] Step 131: Take each node in the user similarity relationship network as a community and calculate the modularity gain between the first node and the community where each other node is located for the i-th time, where the first node is any one of all nodes and the initial value of i is 1.

[0103] In the embodiment of the present invention, each node in the user similarity relationship network is regarded as an independent community, and the modularity gain is calculated for all nodes. For example, the community where node u (the first node) is located is c u , the community where node v (any other node) is located is c v , then, node u and community c v The modular gain formula is:

[0104]

[0105] Where, ∑ in Indicates community c v The sum of the edge weights within k u,in Indicates the link from node u to community c v The sum of the weights of the nodes in ∑ tot Indicates a link to the community c v The sum of the edge weights of any node in k u represents the sum of the weights of all edges linked to node u, and m represents the sum of the weights of all edges in the user similarity relationship network.

[0106] The above method can be used to calculate the gain between any node and the community where other nodes are located.

[0107] Step 132: When there is a positive modularity gain between the first node and the community where each other node is located, the first node is moved to the community where the second node is located to obtain an aggregated community, where the second node is the node corresponding to the maximum modularity gain among the modularity gains between the first node and the community where each other node is located.

[0108] The embodiment of the present invention selects a local optimal method to perform the mobile operation. Considering the modularity gain between two adjacent nodes, the node u is moved from the original community c u Move to community c where node v is located v The modularity gain size is used for the movement operation.

[0109] Specifically, if the first node does not generate a positive gain when moving to the community where other nodes are located, it will not move; if the first node has a positive gain when moving to the community where other nodes are located, the first node will move to the community with the maximum modularity gain to obtain the aggregated community.

[0110] Step 133: Take each aggregated community as a new node, with i=i+1, and continue to perform the step of calculating the modularity gain between the first node and the communities where other nodes are located for the i-th time until the preset stopping condition is met, thereby obtaining several high-density communities in the user similarity relationship network.

[0111] In this embodiment of the present invention, each aggregated community is regarded as a new node, and the edge weights between communities are converted into edge weights between new nodes. That is, the edge weight between two new nodes is equal to the sum of the edge weights between the nodes in the two communities, so as to construct a new network.

[0112] At this step, the process jumps to Step 131 and iterates until the user similarity network cannot be simplified any further, or a predetermined stopping condition is reached, such as stopping after a predetermined number of iterations. The user similarity network is ultimately decomposed into multiple high-density communities through the above method.

[0113] In one embodiment of the present invention, step 140, determining the ranking weights of users in a high-density community based on the behavioral structure characteristics of the high-density community, specifically includes:

[0114] Step 141. Calculate the preset collusion index based on the behavioral structural characteristics of the high-density community to obtain the collusion index data of the high-density community.

[0115] In order to eliminate the impact of abnormal data, the behavioral and structural characteristics of colluding users can be considered, and a collusion indicator can be introduced to calculate the weight of high-density communities.

[0116] In one embodiment of the present invention, the collusion index may include ranking deviation, user ratio, and community size. Specifically, Step 141 calculates a preset collusion index based on the behavioral structure characteristics of the high-density community to obtain the collusion index data of the high-density community, which specifically includes:

[0117] B1. Obtain a target ranking object set within a target high-density community and ranking data of the target ranking object set by users within the target high-density community based on the behavioral structural characteristics of the high-density community, wherein the target high-density community is any one of all high-density communities, and the target ranking object set is a set of objects ranked by two or more users within the target high-density community; obtain ranking data of the target ranking object set by communities other than the target high-density community; calculate the deviation based on the ranking data of the target ranking object set within the target high-density community and the ranking data of the target ranking object set by other communities to obtain the ranking deviation of the target high-density community.

[0118] Specifically, the ranking deviation within and outside the target high-density community is calculated based on the behavioral characteristics. The target high-density community is any one of all high-density communities. The ranking information is used to obtain the target ranking object set of users within the target high-density community, as well as the ranking data of the target ranking objects by users within the target high-density community. The target ranking object set is the set of objects ranked by two or more users within the target high-density community.

[0119] Based on the behavioral characteristics of the above-mentioned target high-density community and the behavioral characteristics of other communities outside the target high-density community, the ranking deviation inside and outside the target high-density community is determined.

[0120] In one embodiment of the present invention, the ranking deviation is the average of the average ranking deviations of objects inside and outside the high-density community, expressed as:

[0121]

[0122] Where RD(g) represents the ranking deviation of high-density community g, I g represents the set of target ranking objects ranked by more than two users in the high-density community g, gavg(i) and ngavg(i) represent the average ranking of object i inside and outside the high-density community g, respectively.

[0123] It is understandable that the larger the ranking deviation RD(g) of the high-density community g is, the more likely the high-density community g is to be a collusion user community.

[0124] B2. Based on the behavioral structure characteristics of high-density communities, obtain the number of users within the target high-density community who rank the target ranking object, where the target ranking object is any one of the target ranking object set; obtain the total number of users who rank the target ranking object. Calculate the ratio of the number of users within the target high-density community who rank the target ranking object to the total number of users to obtain the user ratio of the target high-density community.

[0125] Specifically, the ratio of the number of users who rank the object to the total number of users in the high-density community is calculated based on the behavioral characteristics. The number of users who rank the target ranking object in the target high-density community is obtained through ranking information. Based on the behavioral characteristics of the target high-density community and the total number of users who rank the target ranking object in the entire user similarity relationship network, the maximum value of the ratio of the number of users who rank the target ranking object in the high-density community to the total number of users who rank the target ranking object is obtained, and this value is used as the user ratio of the target high-density community, which can be calculated using the following formula:

[0126]

[0127] Where UR(g) represents the proportion of users in high-density community g, U i Indicates the total number of users who rank the target ranking object i, U gi It represents the number of users who rank the target ranking object i in the high-density community g.

[0128] It can be understood that the larger the user ratio UR(g) of the high-density community g is, the greater the possibility that the high-density community g is a collusion user community.

[0129] B3. Determine the community size of the target high-density community based on the behavioral structural characteristics of the high-density community.

[0130] Specifically, the number of users in each high-density community is obtained through ranking information, and the community size GS(g) of the high-density community is obtained:

[0131]

[0132] Where U g represents the number of users in the high-density community g.

[0133] It can be understood that the larger the community size GS(g) of the high-density community g is, the greater the possibility that the high-density community g is a collusion user community.

[0134] B4. The ranking deviation, user ratio and community size of the target high-density community constitute the collusion indicator data of the target high-density community.

[0135] Step 142: Calculate the ranking weight of each user in the high-density community based on the collusion index data.

[0136] In one embodiment of the present invention, the ranking weight of each user in a high-density community is calculated using the following formula:

[0137]

[0138] Where W(g) is the ranking weight of the target high-density community, RD(g) is the ranking deviation of the target high-density community, UR(g) is the user ratio of the target high-density community, GS(g) is the community size of the target high-density community, and U g is the number of users in the target high-density community.

[0139] The embodiment of the present invention formally describes ranking data through network science methods, establishes overall associations between users, and obtains a corresponding user similarity relationship network; identifies high-density communities in the user similarity relationship network through community mining, that is, suspected groups of collusion attacks, calculates the ranking weights of high-density communities, and then aggregates the ranking results of multiple users based on the ranking weights, which can effectively eliminate the impact of abnormal data such as malicious reviews and collusion attacks.

[0140] In one embodiment of the present invention, assume that eight users rank five objects, with each ranking including three objects. The specific rankings are shown in Table 1. As can be seen from Table 1, the rankings of users u6, u7, and u8 are identical and significantly inconsistent with the rankings of the other judges, suggesting a conspiracy attack.

[0141] Table 1 Ranking matrix

[0142] <![CDATA[i1]]> <![CDATA[i2]]> <![CDATA[i3]]> <![CDATA[i4]]> <![CDATA[i5]]> <![CDATA[u1]]> 1 2 3 0 0 <![CDATA[u2]]> 1 2 0 3 0 <![CDATA[u3]]> 0 1 2 3 0 <![CDATA[u4]]> 1 2 0 0 3 <![CDATA[u5]]> 1 0 2 3 0 <![CDATA[u6]]> 3 2 1 0 0 <![CDATA[u7]]> 3 2 1 0 0 <![CDATA[u8]]> 3 2 1 0 0

[0143] The ranking aggregation method of the present invention is used to aggregate the ranking, and the threshold β = 0.6 is selected. First, the overall association between users is established, and the corresponding user similarity relationship network is obtained, such as Figure 2 As shown, Figure 2 The user similarity relationship network in the embodiment of the present invention. Secondly, the Louvain community detection algorithm is used to identify high-density communities in the user similarity relationship network. Figure 2 It can be seen that users u6, u7, and u8 are identified as suspected collusion groups. Based on this, the collusion index is used to calculate the weight of each user, as shown in Table 2. Finally, combining these weights with user ranking data, the aggregate ranking is obtained using the Borda counting method, as shown in Table 3. The ranking results are: v1 > v2 > v3 > v4 > v5. This shows that the ranking aggregation method proposed in this paper can effectively eliminate the impact of abnormal data.

[0144] Table 2 User weights

[0145] <![CDATA[w1]]> <![CDATA[w2]]> <![CDATA[w3]]> <![CDATA[w4]]> <![CDATA[w5]]> <![CDATA[w6]]> <![CDATA[w7]]> <![CDATA[w8]]> 1 1 1 1 1 0.4656 0.4656 0.4656

[0146] Table 3 Weighted ranking matrix

[0147] <![CDATA[i1]]> <![CDATA[i2]]> <![CDATA[i3]]> <![CDATA[i4]]> <![CDATA[i5]]> <![CDATA[u1]]> 1 2 3 0 0 <![CDATA[u2]]> 1 2 0 3 0 <![CDATA[u3]]> 0 1 2 3 0 <![CDATA[u4]]> 1 2 0 0 3 <![CDATA[u5]]> 1 0 2 3 0 <![CDATA[u6]]> 1.3695 0.9312 0.4656 0 0 <![CDATA[u7]]> 1.3695 0.9312 0.4656 0 0 <![CDATA[u8]]> 1.3695 0.9312 0.4656 0 0

[0148] The embodiment of the present invention also proposes a ranking aggregation device, which can be referred to Figure 3 , Figure 3 This is a ranking aggregation device in an embodiment of the present invention, and the device includes: an information acquisition unit 301 , a network decomposition unit 302 and a ranking aggregation unit 303 .

[0149] The information acquisition unit 301 is configured to acquire ranking information, where the ranking information includes a ranking result of each user on a preset object.

[0150] The network decomposition unit 302 is configured to construct a user similarity relationship network based on the ranking information, and decompose the user similarity relationship network into a plurality of high-density communities according to a preset community detection algorithm.

[0151] The ranking aggregation unit 303 is configured to obtain the behavioral structure characteristics of the high-density community based on the ranking information, and determine the ranking weights of users in the high-density community according to the behavioral structure characteristics of the high-density community.

[0152] The ranking aggregation result of the object is calculated based on the preset aggregation method using the ranking weight and ranking information.

[0153] The ranking aggregation device proposed in the embodiment of the present invention establishes an overall association between users based on their ranking results and obtains a corresponding user similarity relationship network; through community mining, high-density communities in the user similarity relationship network, that is, suspected groups of collusion attacks, are identified, the ranking weight of each community is calculated, and the ranking results of multiple users are aggregated based on the ranking weights, which can effectively eliminate the influence of abnormal data such as malicious evaluations and collusion attacks, making the aggregated ranking results more accurate and reliable.

[0154] Figure 4 FIG1 shows the internal structure of a computer device in one embodiment of the present invention. The computer device can be a terminal or a system. Figure 4 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When the computer program is executed by the processor, the processor can implement the various steps in the above method embodiment. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor can implement the various steps in the above method embodiment. It will be understood by those skilled in the art that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0155] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes each step in the above method embodiment.

[0156] In one embodiment, a computer-readable storage medium is proposed, which stores a computer program. When the computer program is executed by a processor, the processor executes the various steps in the above-mentioned method embodiment. It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0157] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0158] The above embodiments merely illustrate several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A ranking aggregation method, characterized in that: The method comprises: Obtaining ranking information, wherein the ranking information includes a ranking result of each user for a preset object; Constructing a user similarity relationship network based on the ranking information, and decomposing the user similarity relationship network into a plurality of high-density communities according to a preset community detection algorithm; Obtaining behavioral structural characteristics of the high-density community based on the ranking information, and determining ranking weights of users in the high-density community according to the behavioral structural characteristics of the high-density community; Obtaining a ranking aggregation result of the object by calculating the ranking weight and the ranking information based on a preset ranking aggregation method; The user similarity relationship network is decomposed into a number of high-density communities according to a preset community detection algorithm, specifically including: Taking each node in the user similarity relationship network as a community, calculating the modularity gain between the first node and the community where each other node is located for the i-th time, where the first node is any one of all nodes and the initial value of i is 1; When there is a positive modularity gain between the first node and the communities where each other node is located, the first node is moved to the community where the second node is located to obtain an aggregated community, wherein the second node is the node corresponding to the maximum modularity gain among the modularity gains between the first node and the communities where each other node is located; Each of the aggregated communities is treated as a new node, with i=i+1, and the step of calculating the modularity gain between the first node and the communities where other nodes are located is continued for the i-th time until a preset stopping condition is met, thereby obtaining a plurality of high-density communities in the user similarity relationship network; The step of determining the ranking weights of users in the high-density community according to the behavioral structure characteristics of the high-density community specifically includes: Calculating a preset collusion index according to the behavioral structural characteristics of the high-density community to obtain collusion index data of the high-density community; The ranking weight of each user in the high-density community is calculated based on the collusion index data.

2. The method according to claim 1, wherein The step of constructing a user similarity relationship network based on the ranking information specifically includes: Normalizing the ranking information to obtain ranking data of the object by each user; Calculating user similarity between two users based on the ranking data of the objects by each user, wherein the user similarity is the similarity between the ranking results of two users; Use users as nodes and similarities as edges to build a user similarity relationship network.

3. The method according to claim 2, wherein The calculating the user similarity between two users based on the ranking data of the objects by each user specifically includes: Obtain a first ranking object set of a first user and a second ranking object set of a second user, wherein the first user and the second user are any one of all users; taking the union of the first ranking object set and the second ranking object set as a third ranking object set; Calculate, based on the ranking data of the objects by the respective users and the third ranking object set, the ranking similarity between the first user and the second user for each ranking object in the third ranking object set; The user similarity between the two users is calculated based on the ranking similarity of the first user and the second user to each ranking object in the third ranking object set.

4. The method according to claim 1, wherein The step of calculating a preset collusion index based on the behavioral structural characteristics of the high-density community to obtain the collusion index data of the high-density community specifically includes: Obtaining a target ranking object set within a target high-density community according to the behavioral structure characteristics of the high-density community, and ranking data of the target ranking objects combined by users within the target high-density community, wherein the target high-density community is any one of all high-density communities, and the target ranking object set is a set of objects ranked by two or more users within the target high-density community; Obtaining ranking data of the target ranking object set from other communities except the target high-density community; Calculating the deviation of the ranking of the target high-density community based on the ranking data of the target ranking object set in the target high-density community and the ranking data of the target ranking object set in other communities to obtain the ranking deviation of the target high-density community; obtaining the number of users who rank a target ranking object in the target high-density community based on the behavioral structure characteristics of the high-density community, where the target ranking object is any one of the target ranking object set; Obtaining the total number of users who have ranked the target ranking object; Calculate a ratio based on the number of users who rank the target ranking object in the target high-density community and the total number of users to obtain a user ratio of the target high-density community; Determining the community size of the target high-density community according to the behavioral structural characteristics of the high-density community; The ranking deviation, user ratio and community size of the target high-density community constitute the collusion indicator data of the target high-density community.

5. The method according to claim 4, wherein The ranking weight of each user in the high-density community is calculated by the following formula: Where W(g) is the ranking weight of the target high-density community, RD(g) is the ranking deviation of the target high-density community, UR(g) is the user ratio of the target high-density community, GS(g) is the community size of the target high-density community, and U g is the number of users in the target high-density community.

6. A ranking aggregation device, characterized in that: The device comprises: an information acquisition unit, a network decomposition unit and the ranking aggregation unit; The information acquisition unit is used to acquire ranking information, wherein the ranking information includes the ranking result of each user for the preset object; The network decomposition unit is used to construct a user similarity relationship network based on the ranking information, and decompose the user similarity relationship network into a plurality of high-density communities according to a preset community detection algorithm; The ranking aggregation unit is configured to obtain the behavioral structure characteristics of the high-density community based on the ranking information, and determine the ranking weights of users in the high-density community according to the behavioral structure characteristics of the high-density community; Obtaining a ranking aggregation result of the object by calculating the ranking weight and the ranking information based on a preset ranking aggregation method; The network decomposition unit is further configured to treat each node in the user similarity relationship network as a community and calculate the modularity gain between the first node and the community where each other node is located for the i-th time, wherein the first node is any one of all nodes and the initial value of i is 1; When there is a positive modularity gain between the first node and the communities where each other node is located, the first node is moved to the community where the second node is located to obtain an aggregated community, wherein the second node is the node corresponding to the maximum modularity gain among the modularity gains between the first node and the communities where each other node is located; Each of the aggregated communities is treated as a new node, with i=i+1, and the step of calculating the modularity gain between the first node and the communities where other nodes are located is continued for the i-th time until a preset stopping condition is met, thereby obtaining a plurality of high-density communities in the user similarity relationship network; The ranking aggregation unit is further configured to calculate a preset collusion index based on the behavioral structure characteristics of the high-density community to obtain collusion index data of the high-density community; The ranking weight of each user in the high-density community is calculated based on the collusion index data.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 5.

8. A computer device comprising a memory and a processor, characterized in that: The memory stores a computer program, and when the computer program is executed by the processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 5.

Images