A network user public opinion portrait construction method in an anonymous environment

Abstract

The application discloses a network user public opinion portrait construction method in an anonymous environment, comprising: real anonymous user feature acquisition, real anonymous user equivalence class construction, and new user portrait identification. The application reduces the user portrait space by establishing an equivalence relationship and constructing an equivalence class, solves the problems of low portrait search efficiency and large redundant anonymous user portraits occupying the portrait library in the anonymous environment, and thus realizes quick matching of the new user portrait with a small space cost, and can efficiently construct a credibility label for the real anonymous user comment and provide assistance for public opinion management.

Description

Technical Field

[0001] This invention belongs to the field of cyberspace security and relates to a method for constructing online user opinion profiles in an anonymous environment. Background Technology

[0002] Social networking platforms and other social media have rapidly penetrated people's lives, attracting hundreds of millions of users due to their fast information dissemination, wide sharing range, strong timeliness, and high interactivity. To protect user privacy, anonymity has become an important channel for users to express their opinions on sensitive topics. However, anonymity also brings difficulties to the governance of online public opinion. A single user can publish multiple anonymous comments, making it difficult to trace these comments back to the same user. Therefore, using traditional methods of constructing public opinion profiles leads to a dramatic expansion of the profile database. Furthermore, because online public opinion is dynamic and real-time, the speed at which user profiles are generated determines whether public opinion regulators can promptly assess new users under a given topic. For hot topics with a large number of users, frequent reconstruction of user profiles is too costly and lacks timeliness. The excessive size of the profile database due to redundant anonymous user profiles also makes rapid profile construction through matching the database inefficient. In addition, malicious use of anonymity can achieve the effect of online trolls, creating a severely negative influence on public opinion. At critical moments, combining anonymity with online trolls can form a public opinion attack, damaging the target's public image and interfering with their normal actions. Summary of the Invention

[0003] To address the aforementioned issues, this invention proposes a label-based equivalence class construction method, centered on the equivalence class approach and based on the idea of ​​"grouping people by their kind." This method constructs user public opinion profiles in an anonymous environment, achieving the reduction of massive user profiles and improving the efficiency of constructing new user profiles, thereby providing strong data support for decision-making in online public opinion governance.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A method for constructing online user opinion profiles in an anonymous environment includes the following steps:

[0006] Step 1: Obtain anonymous user characteristics;

[0007] Step 2: Construct the equivalence class for anonymous users;

[0008] Step 3: New user profile identification.

[0009] Furthermore, step 1 specifically includes the following sub-steps:

[0010] (1) For anonymous users under a topic, their features are vector C. A=C1, where C1:=(a1, a2, a3, a4, a5, L1), a1 refers to whether the comment IP is consistent with the location of the topic; a2 is the difference between the comment time and the topic appearance time; a3 is the user sentiment, divided into radical and conservative; a4 is the stance of the statement, divided into support and opposition; a5 refers to setting a keyword list for text matching, the feature value is whether the comment text hits the keyword; L1 is the comment tag, which is matched with the typical malicious comment list based on the text content, divided into normal and malicious;

[0011] (2) Real-name users under the topic are characterized by triple C. R = (C1, C2, C3), where C2: = (b1, b2, b3, b4, b5, b6, b7), b1 represents whether the user is likely to use a proxy, b2 represents the number of followers, b3 represents the number of fans, b4 represents the number of likes, b5 represents the number of favorites, b6 represents the number of likes, and b7 represents the number of related historical comments posted by the user; The matrix element a nk L represents the k-th feature of the n-th related historical comment of a verified user, and its extraction method is consistent with the k-th feature of C1; n1 This is the tag for the nth related historical comment.

[0012] Furthermore, step 2 specifically includes the following sub-steps:

[0013] Step 2.1: Constructing the user vectorized feature point set;

[0014] Step 2.2: Description of equivalence relations between users;

[0015] Step 2.3: Construct the user equivalence class;

[0016] Step 2.4: Simplify the user profile space.

[0017] Further, step 2.1 includes the following process:

[0018] Construct Set = {c A1 c A2 c Ak c R1 c R2 c Rr}, where for the k-th anonymous user, according to C A c Ak =(a1, a2, a3, a4, a5), L Ak =L1; For the r-th real-name user, from C R Feature c extracted from tuple C1 Rr =(a1, a2, a3, a4, a5), L Rr=L1; All binary features are quantized to 0 and 1, and a2 is normalized in days.

[0019] Furthermore, step 2.3 includes the following process:

[0020] (1) Input the number of equivalent centers K, select a feature point from Set as the initial center Z1, and continuously select the point with the largest minimum distance from the selected center as Z1. i This continues until K initial centers are selected;

[0021] (2) Select a point c from the feature point set Set that has not been added to the equivalence class. If min{||cZ i (k)||,i=1,2,…,K}=||cZ j (k)||, then c∈S j (k); where j is the equivalence class number and k is the iteration number;

[0022] (3) Calculation Where c i It is the point corresponding to label L=0, c j α represents the number of points corresponding to label L=1, m and n are the number of points corresponding to L=0 and L=1 respectively, and α and β satisfy the condition: α >> β;

[0023] (4) Repeat steps (2) and (3) until Z is reached. j (k+1)=Z j (k);

[0024] (5) Calculate the equivalence class S j Class tags Where m and n are the number of points L = 0 and 1 respectively, and α and β satisfy the condition: α >> β;

[0025] (6) Output equivalence classes S1, S2…S j Equivalence centers Z1, Z2, ..., Z j and class tags

[0026] Furthermore, step 2.4 includes the following process:

[0027] If the anonymous user clicks c A ∈S j Then it is simplified to the same anonymous portrait P Aj This includes the comment preference vector M and the user credibility T, where M = Z j , Let c represent the equivalence class feature and equivalence class label, respectively; the anonymous user profile space is reduced to a set of K profiles; if a real-name user clicks c R ∈S jIf the feature matrix C3 is not empty, then feature correction is performed to obtain the independent user profile P. R Including the comment preference vector M R User credibility T R and influence characteristic C2, where M R and T R The calculation method is as follows:

[0028] (1) According to C3, let c i =(a i1 a i2 a i3 a i4 a i5 ), 1≤i≤n; calculate Where L i1 α is the tag for the i-th related historical comment. i β i Meet the conditions

[0029] (2) Calculation Where c = Z j , where p is the length of the vector c. For comment tags;

[0030] (3) Calculate M R =JG1+(1-J)G2, where

[0031] (4) Calculation

[0032] If the feature matrix C3 is empty, then it is reduced to the same real-name portrait P. Rj It includes the comment preference vector M, user credibility T, and simplified influence feature C′, where M = Z j , s is the real-name user point c where C3 is empty. R Quantity, C 2i Let be the influence feature of the i-th user.

[0033] Furthermore, step 3 specifically includes the following sub-steps:

[0034] Step 3.1: User feature initialization;

[0035] Step 3.2: Equivalence relation search based on reduced image space;

[0036] Step 3.3: Building a new user profile.

[0037] Further, step 3.1 includes the following process:

[0038] If the new user is anonymous, extract its features as vector c from C1. A = (a1, a2, a3, a4, a5); If the new user is a verified user, extract its features as vector c. R = (C, C2), where feature C is calculated as follows:

[0039] Based on C1 and C3, let c = (a1, a2, a3, a4, a5), c i =(a i1 a i2 a i3 a i4 a i5 ), 1≤i≤n; calculate Where n is the number of rows in matrix C3.

[0040] Furthermore, step 3.2 includes the following process:

[0041] (1) Define the reduced image space SP = {P A1 , ..., P AK P R1 , ..., P RK , ..., P RK+V}, where K is the number of equivalence classes and V is the number of real-name user points in feature matrix C3 that are not empty;

[0042] (2) If the new user is an anonymous user, further reduce the profile space to SP1 = {P A1 , ..., P AK};If min{||c A -M Ai ||,i=1,2,…,K}=||c A -M Aj ||, then the new user profile P new With P Aj equivalence;

[0043] (3) If the new user is a real-name user and C3 is empty, then further simplify the profile space to SP2 = {P R1 , ..., P RK If min{||CM Ri ||,i=1,2,…,K}=||CM Rj ||, then the new user profile P new With P Rj Equivalent; if C3 is not empty, then further reduce the image space to SP3 = {P RK+1 , ..., P RK+V};If min{||CM RK+i||,i=1,2,…,V}=||CM RK+v ||, then the new user profile P new With P RK+v equivalence.

[0044] Furthermore, step 3.3 includes the following process:

[0045] If P new With P Aj or P Rk Equivalent, where 1≤k≤K, then P new Simplify to P Aj or P Rk No new portraits will be added; if P new With P RK+v Equivalent, where 1≤v≤y, then P new Add simplified portrait space SP, P new Including comment preference vector C, user credibility T RK+v And influence characteristic C2.

[0046] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0047] (1) By establishing an equivalence relation and constructing equivalence classes to simplify the user profile space, the problem of a large number of redundant anonymous user profiles occupying the profile library and low profile search efficiency in the anonymous environment is solved. Thus, new user profiles can be quickly matched with a small space cost, and credibility tags can be efficiently constructed for anonymous user comments, providing assistance for public opinion governance.

[0048] (2) The method of this invention fills the gap in previous studies that did not pay enough attention to the anonymous public opinion environment. This method reduces the user profile space by using point set-based equivalence class features and class labels, simplifying a large number of users into a small number of typical user profiles, thereby reducing the size of the user profile database and solving the problem of profile space expansion caused by the anonymity mechanism.

[0049] (3) This invention can quickly build profiles for newly added users by simplifying the profile space and update the profile library, thereby improving the efficiency of new user profile construction. At the same time, the simplified profile space is updated to a certain extent, which fills the vacuum period between two public opinion profile reconstructions and ensures the real-time nature of public opinion analysis, thereby supporting public opinion regulators in controlling public opinion trends and evaluating commenting users. Attached Figure Description

[0050] Figure 1 The flowchart illustrates a method for constructing online user opinion profiles in an anonymous environment, as provided by this invention.

[0051] Figure 2 A model for extracting network user features.

[0052] Figure 3 Construct an algorithm flowchart for user equivalence classes.

[0053] Figure 4 A simplified flowchart for creating user profiles.

[0054] Figure 5 Example of creating an independent profile for real-name users.

[0055] Figure 6 This is a typical example of an anonymous user profile. Detailed Implementation

[0056] The technical solutions provided by the present invention will be described in detail below with reference to specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0057] This invention provides a method for constructing online user opinion profiles in an anonymous environment, the process of which is as follows: Figure 1 As shown, the specific steps include:

[0058] Step 1, Obtaining Anonymous User Features

[0059] Write a Python web crawler to scrape anonymous user comments and verified user IDs, nicknames, genders, and their historical information under a target topic on Zhihu. Figure 2As shown. The obtained dataset is preprocessed to remove advertising information and useless data. A list List1 with 7 columns is created to store the C1 of real and anonymous users: = (a1, a2, a3, a4, a5, L1). Column 1 is the user ID, starting from 1 for real users and incrementing sequentially, while anonymous users are all 0. Columns 2 to 7 are the current comment feature C1 six-tuples: a1 indicates whether the comment IP is consistent with the topic's location; if the user's IP is the same as the region involved in the current topic, it is set to 1, otherwise it is set to 0; a2 is the difference between the comment's posting time and the topic's posting time, which needs to be normalized and retained to two decimal places; a3 is the user's sentiment, set to 1 for radical and 0 for conservative; a4 is the user's stance, set to 1 for support and 0 for opposition; a5 is the number of keywords, matched using a preset keyword table; if the comment text matches the keyword, it is set to 1, otherwise it is set to 0. L1 is the tag for this comment, constructed using the matching results of a pre-configured list of typical malicious comments, MaList. If the comment text matches MaList, it is set to 0; otherwise, it is set to 1. For verified users under the topic, the influence feature vector C2 and feature matrix C3 are extracted. A list List2 with 8 columns is created to store the influence feature vector C2 of verified users. The first column is the user ID, which matches List1; columns 2 to 8 are C2:=(b1, b2, b3, b4, b5, b6, b7). The second column b1 indicates whether a proxy is used; it is set to 1 if a proxy is used, and 0 otherwise. The third column b2 is the number of followers the user follows; the fourth column b3 is the number of fans; the fifth column b4 is the number of likes; the sixth column b5 is the number of favorites; the seventh column b6 is the number of likes; and the eighth column b7 is the number of related historical comments posted by the user, i.e., the total number of comments related to the target topic. Create a feature matrix C3. For each verified user, process their relevant historical comments into six-tuples and store them in a single row of feature matrix C3. Each matrix corresponds to a user ID. The matrix element a nk L represents the k-th feature of the n-th related historical comment of a verified user, and its extraction method is consistent with the k-th feature of C1; n1 This is the tag of the nth related historical comment. The anonymous user's features are represented by vector C. A =C1, the characteristic of a real-name user is the triple G. R = (C1, C2, C3).

[0060] Step 2, construct the equivalence class for anonymous users.

[0061] 2.1 Construction of User Vectorized Feature Point Set

[0062] Construct Set = {c A1 c A2 c Ak c R1c R2 c Rr}, where for each anonymous user, extract c from List1 Ak =(a1, a2, a3, a4, a5), L Ak =L1, c Ak Store in a Set and save it along with the tag L Ak The correspondence; for each real-name user, extract feature c from List1. Rr =(a1, a2, a3, a4, a5), L Rr =L1, similarly, c Rr Store in a Set and save it along with the tag L Ak The correspondence is as follows. All binary features are quantized into 0 and 1, and a2 is normalized in days.

[0063] 2.2 Description of Equivalence Relations Between Users

[0064] For two feature points c1 and c2 in a user vectorized feature point set Set, if Center(c1) = Center(c2), then c1 and c2 have an equivalence relation R, where Center(c) represents the closest equivalence center of the user feature point c.

[0065] 2.3 Constructing User Equivalence Classes

[0066] (1) Input the number of equivalent centers K, select a feature point from Set as the initial center Z1, and continuously select the point with the largest minimum distance from the selected center as Z1. i This continues until K initial centers are selected.

[0067] (2) Select a point c from the feature point set Set that has not been added to the equivalence class. If min{||cZ i (k)||,i=1,2,…,K}=||cZ j (k)||, then c∈S j (k). Where j is the equivalence class number and k is the iteration number.

[0068] (3) Calculation Where c i It is the point corresponding to label L=0, c j Let L be the number of points corresponding to label L=1, m and n be the number of points corresponding to L=0 and L=1 respectively, and let α and β satisfy the condition: α >> β.

[0069] (4) Repeat steps (2) and (3) until Z is reached. j (k+1)=Z j (k).

[0070] (5) Calculate the equivalence class S jClass tags Where m and n are the number of points L = 0 and 1 respectively, and α and β satisfy the condition: α >> β.

[0071] (6) Output equivalence classes S1, S2…S j Equivalence centers Z1, Z2, ..., Z j and class tags

[0072] This example demonstrates how to write a Python program to implement the user equivalence class construction algorithm, and its flowchart is shown below. Figure 3 As shown. The algorithm takes a user feature point set `Set` and its corresponding labels as input, and sets weights α = 0.9 and β = 0.1. Starting from K = 2, the weights are incremented, and the algorithm constructs equivalence classes by inputting the parameters and calculating the sum of squared errors. Where p is the set of equivalence class points S j For the feature points in the dataset, select the value of K where the trend of SSE decreases sharply as K increases, and use this value as the number of equivalence classes. Output the set of K equivalence class points S. j and the equivalent center Z corresponding to each point set j and class tags

[0073] 2.4 Simplifying User Profile Space

[0074] like Figure 4 As shown, the set S of K equivalence class points j and the equivalent center Z corresponding to each point set j and class tags As input, for each equivalence class point set S j Reduce all anonymous users to the same anonymous profile P. Aj ,like Figure 6 As shown, it includes a comment preference vector M and a user credibility T, where M = Z j , Let P represent the equivalence class feature and equivalence class label, respectively. The anonymous user profile space is reduced to a set of K profiles. Let P be the K anonymous profiles. Aj Store the profile in the user profile space SP, saving the set of anonymous users corresponding to that profile. From S j Retrieve the real-name user IDs and their corresponding C2 and C3 from List1 and List2. If the feature matrix C3 of the real-name user is empty, reduce it to the same real-name profile P. Rj It includes the comment preference vector M, user credibility T, and simplified influence feature C′, where M = Z j , Write a function to calculate C′. s is the real-name user point c where C3 is empty. R Quantity, C 2iLet C' be the influence feature of the i-th user. In this example, the input is the influence feature vector C2 of s real-name users whose C3 is empty, and the output is the simplified influence feature C'. Write a Python function. Implement the real-name user M in step M2.4 R and T R The calculation method involves K anonymous portraits P. Rj Store the profile in the user profile space SP, saving the set of real-name users corresponding to that profile. If the feature matrix C3 of the real-name users is not empty, then call the function. Perform feature correction and output M R and T R Build independent user profiles P R ,like Figure 5 As shown, it includes the comment preference vector M R User credibility T R and influence characteristics C2. M R and T R The calculation method is as follows:

[0075] (1) According to C3, let c i =(a i1 a i2 a i3 a i4 a i5 ), 1≤i≤n. Calculate Where L i1 α is the tag for the i-th related historical comment. i β i Meet the conditions

[0076] (2) Calculation Where c = Z j , where p is the length of the vector c. For comment tags.

[0077] (3) Calculate M R =JG1+(1-J)G2, where

[0078] (4) Calculation

[0079] The profile of each remaining real-name user P R The data is stored in the user profile space SP. After profile space reduction, in this embodiment, only 261 profiles were constructed for 1003 real anonymous users, and the profile space was reduced to 26.02% of the original space. For larger datasets, the benefits of profile space reduction are more obvious, and it can effectively solve the problem of profile space expansion caused by anonymization mechanisms.

[0080] Step 3, New User Profile Identification Method

[0081] 3.1 User Feature Initialization

[0082] If the new user is anonymous, extract its features as vector c from C1. A = (a1, a2, a3, a4, a5), with its index 0, and the original label is empty; write a Python function CAL_C(C1, C3) to calculate feature C in M3.1. If the new user is a real-name user, extract the feature triplet (C1, C2, C3), call CAL_C(C1, C3), output the feature vector C, and initialize the real-name user feature as c. R = (C, C2), whose number is the last real-name user ID + 1, and the original label is empty. The calculation method for feature C is as follows:

[0083] Based on C1 and C3, let c = (a1, a2, a3, a4, a5), c i =(a i1 a i2 a i3 a i4 a i5 ), 1≤i≤n. Calculate Where n is the number of rows in matrix C3.

[0084] 3.2 Equivalence relation search based on reduced image space

[0085] The reduced image space SP = {P} output from step 2.3 A1 , ..., P AK P R1 , ..., P RK , ..., P RK+V} (where K is the number of equivalence classes and y is the number of real-name user points that are not empty in the feature matrix C3) extracts the subspace SP1 = {P} based on user type. A1 , ..., P AK}, SP2={P R1 , ..., P RK} and SP3 = {P RK+1 , ..., P RK+V Write a function SPSearch(SP, c) to perform the user equivalence relation search in step 3.2, returning a profile equivalent to the new user. If the new user is anonymous, call SPSearch(SP1, c) A If min{||c A -M Ai ||,i=1,2,…,K}=||c A -MAj ||, then output the equivalent image P Aj If the new user is a verified user and C3 is empty, then call SPSearch(SP2, c...). R If min{||CM Ri ||,i=1,2,…,K}=||CM Rj ||, Output equivalent image P Rj If C3 is not empty, then call SPSearch(SP3, c R If min{||CM RK+i ||,i=1,2,…,V}=||CM RK+v ||, Output equivalent image P RK+v Because different sized subspaces are extracted for searching different types of users, the time spent searching for equivalence relations of new users is much less than that spent reconstructing user profiles, which improves the efficiency of profile construction and, to some extent, fills the gap in timely evaluation of new users added between two clustering sessions.

[0086] 3.3 New User Profile Construction

[0087] Based on the classification of the equivalent profiles output in step 3.2 in the reduced profile space SP, if the new user profile P new With P Aj or P Rk Equivalent, where 1≤k≤K, then P new Simplify to P Aj or P Rk No new profile will be created; the user will only be added to the P profile. Aj or P Rk The corresponding user set; if P new With P RK+v Equivalent, where 1≤v≤V, construct P for this user. new This includes the comment preference vector C and the user credibility T. RK+v And influence characteristic C2, P new Add the profile to the SP (Special Profile Space), save the unique real-name user ID corresponding to the profile, and complete the profile database update.

[0088] The technical means disclosed in this invention are not limited to those disclosed in the above embodiments, but also include technical solutions composed of any combination of the above technical features. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications are also considered within the scope of protection of this invention.

Claims

1. A method for constructing online user opinion profiles in an anonymous environment, characterized in that, Includes the following steps: Step 1: Obtain anonymous user characteristics; specifically including the following sub-steps: (1) Anonymous users in the topic, Its characteristic is a vector ,in , This refers to whether the IP address of the comment matches the location of the topic. It is the difference between the time of the comment and the time the topic appeared; It refers to user sentiment, which can be divided into radical and conservative. It refers to one's stance, divided into support and opposition; This refers to setting up a keyword list for text matching, with the feature value being whether the comment text matches the keywords. These are comment tags that match text content against a list of typical malicious comments, categorizing them as normal or malicious. (2) Real-name users under the topic, Its characteristic is a triplet ,in , Indicates whether a user is likely to use a proxy. Indicates the number of users following. Indicates the number of followers. This indicates the number of likes received. This indicates the number of times something has been added to favorites. Indicates the number of people who are liked. This indicates the number of related historical comments posted by the user; among which The matrix elements Indicates the number of real-name users The first related historical comment Each feature, its extraction method and The All features are consistent; It is the first Tags related to historical comments; Step 2: Construct the equivalence class for anonymous users; specifically including the following sub-steps: Step 2.1: Constructing the user-vectorized feature point set; specifically including the following process: structure , among which for the first An anonymous user, according to , , For the first A real-name user, from tuple Feature extraction , All binary features are quantized to 0 and 1. Based on days and normalized Step 2.2: Description of equivalence relations between users; Step 2.3: Constructing the user equivalence class; including the following process: (1) Input the number of equivalence centers ,from Choose a feature point as the initial center Continuously select the point with the largest minimum distance from the already selected center as... until the choice One initial center; (2) From the feature point set Select a point that is not included in the equivalence class. ,like ,but ;in For equivalence class number, This is the iteration number; (3) Calculation ,in It is a tag Corresponding points It is a tag Corresponding points and They are and The number of points, and As the weight, satisfying the condition: ; (4) Repeat (2) and (3) until ; (5) Calculate equivalence classes Class tags ,in and They are and The number of points, and As the weight, satisfying the condition: ; (6) Output equivalence classes Equivalence center and class tags Step 2.4: User profile space reduction; specifically including the following process: If anonymous user clicks This can be simplified to the same anonymous portrait. Including comment preference vectors and user credibility ,in , , representing equivalence class features and equivalence class labels respectively; the anonymous user profile space is simplified to A collection of profiles; if a real-name user clicks... And the characteristic matrix If the value is not empty, feature correction is performed to obtain an independent user profile. Including comment preference vectors User credibility and influence characteristics ,in, and The calculation method is as follows: (1) According to ,make 1 ;calculate , ,in For the first Tags related to historical comments, , Meet the conditions ; (2) Calculation ,in , for The length of the vector. For comment tags; (3) Calculation ,in , ; (4) Calculation ; If the characteristic matrix If empty, reduce to the same real-name portrait. Including comment preference vectors User credibility and simplify influence characteristics ,in , , , for Empty real-name user points quantity, For the first Influence characteristics of an individual user; Step 3: New user profile identification.

2. The method for constructing online user opinion profiles in an anonymous environment according to claim 1, characterized in that, Step 3 specifically includes the following sub-steps: Step 3.1: User feature initialization; Step 3.2: Equivalence relation search based on reduced image space; Step 3.3: Building a new user profile.

3. The method for constructing online user opinion profiles in an anonymous environment according to claim 2, characterized in that, Step 3.1 includes the following process: If the new user is an anonymous user, from Extract its features as vectors If the new user is a verified user, extract their features as a vector. Among them, features The calculation method is as follows: according to and ,make , 1 ;calculate ,in For matrix Number of rows.

4. The method for constructing online user opinion profiles in an anonymous environment according to claim 2, characterized in that, Step 3.2 includes the following process: (1) Define the reduced image space ,in The number of equivalence classes, Characteristic matrix The number of real-name registered users that are not empty; (2) If the new user is an anonymous user, further simplify the profile space to ;like Then the new user profile and equivalence; (3) If the new user is a real-name user, and If it is empty, then the image space is further simplified to... ,like Then the new user profile and Equivalent; if If it is not empty, then the image space is further simplified to... ;like Then the new user profile and equivalence.

5. The method for constructing online user opinion profiles in an anonymous environment according to claim 2, characterized in that, Step 3.3 includes the following process: like and or Equivalent, of which Then Simplified or No new portraits will be added; if and Equivalent, of which Then Join the Simple Portrait Space , Including comment preference vectors User credibility and influence characteristics .

Images