User identity linking method and system based on knowledge graph and multi-path attribute fusion

Abstract

The application relates to the technical field of social networks and digital identity recognition, and discloses a user identity linking method and system based on a knowledge graph and multi-path attribute fusion. The method collects information published by a plurality of accounts on different social platforms to form a data set; a semantic knowledge graph based on the accounts is constructed according to the data set; a multi-path attribute combination is set based on all kinds of attribute features, and an attribute combination scheme is determined from the multi-path attribute combination according to the information published by an input account; deep-level attribute features of the input account are extracted, and semantic mapping is performed on the semantic knowledge graph according to the deep-level attribute features, so that a plurality of alternative accounts are screened in the semantic knowledge graph; multi-path attribute combination joint analysis is performed on each alternative account according to the attribute combination scheme, and the similarity between the alternative account and the input account is calculated; and a specified alternative account is selected as a similar account of the input account. The application improves the accuracy and efficiency of user identity recognition.

Description

Technical Field

[0001] This invention relates to the fields of social networks and digital identity recognition technology, specifically a user identity linking method and system based on knowledge graphs and multi-path attribute fusion. Background Technology

[0002] User identity linking is a crucial step in online social network research, aiming to identify different accounts belonging to the same individual. In the context of today's widespread use of social media, users typically create multiple accounts on different platforms to meet their diverse needs; some platforms are primarily used for sharing personal life, while others are mainly used for sharing industry news and social phenomena. The personal information filled out on different platforms may differ, and while there is some overlap in the required fields, there are also certain platform-specific attributes. These factors make cross-platform similar account identification particularly complex and difficult.

[0003] Many existing methods rely on basic user attributes and fail to fully explore and utilize the potential semantic connections between accounts, resulting in suboptimal accuracy and efficiency in identification. Furthermore, differences may exist in personal information submitted across different platforms, and while there is some overlap in the required attributes across platforms, there are also certain platform-specific attributes. These factors make cross-platform similar account identification particularly complex and difficult, thus requiring urgent solutions. Summary of the Invention

[0004] To address the technical problems of poor accuracy and efficiency in existing user identity linking technologies, this invention provides a solution.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] This invention discloses a user identity linking method based on knowledge graph and multi-path attribute fusion, including the following steps, namely S1 to S6.

[0007] S1. Collect information published by several accounts on different social media platforms to form a dataset.

[0008] S2. Construct an account-based semantic knowledge graph based on the dataset; the semantic knowledge graph is used to reflect and store the relationships between various entities; the categories of entities include: user entities, account entities, and attribute feature entities; attribute feature entities include multiple attribute features, each of which is a basic attribute feature or a deep attribute feature.

[0009] S3. Set up multi-path attribute combinations based on all types of attribute features, and determine the attribute combination scheme from the multi-path attribute combinations according to the information published by the input account.

[0010] S4. Extract the deep-level attribute features of the input account, and perform semantic mapping with the semantic knowledge graph accordingly, thereby selecting multiple candidate accounts in the semantic knowledge graph that have an intrinsic relationship with the input account that meets the specified requirements.

[0011] S5. Perform multi-path attribute combination joint analysis on each candidate account according to the attribute combination scheme, and calculate the similarity between each candidate account and the input account.

[0012] S6. Arrange all candidate accounts in descending order of similarity into a queue, and select the first specified number of candidate accounts in the queue as similar accounts to the input account.

[0013] As a further improvement to the above scheme, in step S2, basic attribute features are obtained from the personal information filled in by the user on the social platform; deep attribute features are extracted from the social text formed by the posts and / or comments published by the user on the social platform.

[0014] As a further improvement to the above scheme, in step S2, a fine-tuned large language model is used to perform semantic analysis on social texts, thereby extracting deep-level attribute features.

[0015] As a further improvement to the above scheme, in step S2, the user entity is U, U = {u1, u2, ..., u...} m}; The account entity is A, A={a1,a2,...,a n}; Attribute feature entity F, F={f1,f2,...,f p}; where u m This represents the m-th user; a n f represents the nth account; p Let p represent the p-th attribute feature; m, n, and p are the total number of users, accounts, and attribute features in the semantic knowledge graph, respectively.

[0016] As a further improvement to the above scheme, in step S3, the multi-path attribute combination is a Group, where Group = {g1, g2, ..., g...} q}; where g q This represents the q-th attribute combination; q is the total number of attribute combinations; each attribute combination contains at least two attribute features.

[0017] As a further improvement to the above scheme, the similarity calculation formula in step S5 is as follows:

[0018]

[0019] In the formula, P represents the similarity between the input account and any candidate account; W gLet be the weight value of any attribute combination g; j is any attribute feature in attribute combination g; w j Let be the weight percentage of the j-th attribute feature in the attribute combination g; SIMILAR(·) represents the similarity calculation function; and Let J represent the j-th attribute feature of the input account and any candidate account, respectively.

[0020] As a further improvement to the above scheme, the weight value W g The calculation formula is:

[0021]

[0022] In the formula, Let g and z represent all the attribute features in the attribute combination g, respectively; Reward(·) represents the weight compensation function, which is used to comprehensively evaluate the weight value of the attribute combination based on the number of missing attributes and the number of hit combinations.

[0023] This invention also discloses a user identity linking system based on knowledge graph and multi-attribute fusion, which applies the above-mentioned user identity linking method based on knowledge graph and multi-attribute fusion; the user identity linking system includes: a data acquisition module, a graph construction module, and a data processing module.

[0024] The data acquisition module is used to collect information published by several accounts on different social media platforms to form a dataset;

[0025] The graph construction module is used to construct an account-based semantic knowledge graph based on the dataset; the semantic knowledge graph is used to reflect and store the relationships between various entities; the entity categories include: user entities, account entities, and attribute feature entities; attribute feature entities include multiple attribute features, each of which is a basic attribute feature or a deep attribute feature.

[0026] The data processing module is used to: first, set up multi-attribute combinations based on all types of attribute features, and determine the attribute combination scheme from the multi-attribute combinations according to the information published by the input account; then, extract the deep-level attribute features of the input account, and perform semantic mapping with the semantic knowledge graph to select multiple candidate accounts in the semantic knowledge graph that meet the specified requirements of intrinsic association with the input account; then, perform multi-attribute combination joint analysis on each candidate account according to the attribute combination scheme, and calculate the similarity between each candidate account and the input account; finally, arrange all candidate accounts in descending order of similarity into a queue, and select a specified number of candidate accounts at the top of the queue as similar accounts to the input account.

[0027] Compared with the prior art, the beneficial effects of the present invention are:

[0028] 1. The user identity linking method disclosed in this invention constructs a semantic knowledge graph that not only covers basic account information but also includes deeper indirect information such as topics, opinions, and stances within the social platform. This method considers the inherent connections between attributes and the degree of dispersion across different social platforms, greatly improving the accuracy and efficiency of user identity recognition.

[0029] 2. The multi-path combination scheme joint analysis of this invention can adaptively identify related attribute features, forming general and specific combinations, thus enhancing the accuracy and applicability of the identification process. This method can flexibly adapt to the data characteristics of different social platforms, providing more accurate user links.

[0030] 3. This invention also optimizes the handling of incomplete information. The attribute weight compensation strategy can adjust the weights based on the completeness and contextual relevance of account attributes, effectively reducing the impact of missing attributes on similarity scores. This mechanism not only improves the flexibility of the evaluation process but also enhances its robustness when dealing with diverse and incomplete data.

[0031] 4. This invention, by combining personal information, posts, comments, and other information, can comprehensively analyze and identify user identities. This multi-information fusion strategy effectively solves the limitations that may arise from a single information source and enhances the overall user identification capability.

[0032] 5. This invention utilizes advanced natural language processing technology to achieve efficient mapping and linking of account information. The dynamic attribute compensation mechanism automatically adjusts based on the feature completeness and relevance of the dataset, ensuring the recognition algorithm remains efficient and accurate in constantly changing data environments. Overall, this innovative user identity linking scheme provides an efficient, accurate, and adaptive solution for cross-platform user identification, significantly improving the quality and efficiency of social network analysis and personalized services. Attached Figure Description

[0033] Figure 1 This is a flowchart of the user identity linking method based on knowledge graph and multi-path attribute fusion in Embodiment 1 of the present invention.

[0034] Figure 2 This is an overall structural diagram of the user identity linking method in Embodiment 1 of the present invention (the left side of the diagram shows the relationship between local accounts on two different social platforms, each node represents an account, and the edges represent the relationship between accounts).

[0035] Figure 3 This is an architecture diagram of the user identity linking system based on knowledge graph and multi-path attribute fusion in Embodiment 2 of the present invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] Example 1

[0038] Please see Figure 1 and Figure 2 This invention provides a user identity linking method based on knowledge graph and multi-path attribute fusion. This method can link at least two accounts with the same user identity on at least one social platform. The user identity linking method includes the following steps, namely S1 to S6.

[0039] S1. Collect information published by several accounts on different social media platforms to form a dataset.

[0040] In this embodiment, the information published by accounts on social media platforms includes personal information filled in by users on the social media platforms, as well as posts and comments published by users on the social media platforms. All of this information can be obtained through data crawling.

[0041] It's important to note that comprehensive preprocessing of the crawled data can improve its quality and ensure consistency. This process includes removing duplicate records, correcting errors, and identifying outliers. Simultaneously, data standardization will be performed, such as converting data formats and removing unnecessary emojis, to better suit subsequent analysis procedures. This initial data cleaning and formatting is a crucial step in ensuring the accuracy and efficiency of data analysis.

[0042] S2. Construct an account-based semantic knowledge graph based on the dataset; the semantic knowledge graph is used to reflect and store the relationships between various entities.

[0043] The categories of entities include: user entities, account entities, and attribute feature entities; attribute feature entities include multiple attribute features, each of which is either a basic attribute feature or a deep-level attribute feature.

[0044] The basic attributes are obtained from the personal information that users fill in on social platforms, such as name, address, and gender.

[0045] Deep-level attribute features are extracted from social texts formed by users' posts and / or comments on social platforms, such as user aliases, opinions, and topics.

[0046] It should be noted that there may be similarities, compatibility, contradictions, or irrelevant intrinsic connections between various deep-level attribute characteristics.

[0047] In this embodiment, a fine-tuned large language model can be used to perform semantic analysis on social texts, thereby extracting deep-level attribute features; furthermore, a fine-tuned attribute association model can be used to establish associations between two different attribute features. For example, the following formula can be used:

[0048] Result(att) = LLM1(text)

[0049] R C =LLM2(att) v ,att w )

[0050] In the formula, Result(att) represents the extracted attribute features, such as topics and opinions; LLM1 represents the large language model that needs to be fine-tuned according to the task requirements; and text is the input social text. R C This indicates the intrinsic relationship between two different attributes, such as opinions (att). v and opinions att w Irrelevant; LLM2 is a fine-tuned attribute-related large model.

[0051] In this embodiment, the user entity is U, U = {u1, u2, ..., u}. m}; The account entity is A, A={a1,a2,...,a n}; Attribute feature entity F, F={f1,f2,...,f p}; where u m This represents the m-th user; a n f represents the nth account; p Let p represent the p-th attribute feature; m, n, and p are the total number of users, accounts, and attribute features in the semantic knowledge graph, respectively.

[0052] The following is an example of how relationships between different entities can be represented:

[0053] Such as (u m ,f p )∈R U Indicates user u m Having attribute characteristics f p ;

[0054] For example (a) n ,f p )∈R A Account a n Having attribute characteristics f p ;

[0055] Such as (u m ,a n )∈R UA Indicates user u m With account a n There is a relationship between them, such as a "belongs to" relationship;

[0056] For example (a1,a n )∈R AA This indicates that account a1 and account a n Relationships between them, such as similarity;

[0057] For example (f1, f p )∈R F Representing feature f1 and feature f p They have an inherent relationship, such as a contradictory relationship.

[0058] The semantic knowledge graph constructed by this invention not only includes general information such as account nicknames and avatars, but also richer and deeper indirect attribute features, such as the identity semantics behind nicknames, as well as information such as topics, opinions, and positions within social platforms.

[0059] S3. Set up multi-path attribute combinations based on all types of attribute features, and determine the attribute combination scheme from the multi-path attribute combinations according to the information published by the input account.

[0060] Among them, the multi-way attribute combination is Group, Group = {g1, g2, ..., g q}; where g q This represents the q-th attribute combination; q is the total number of attribute combinations; each attribute combination contains at least two attribute features, for example, g1 = {"Name", "Personal Profile"}. Additionally, the value of q is generally small and can be increased appropriately according to user needs.

[0061] In this embodiment, the attribute combination scheme can be determined by feature engineering screening combined with expert experience.

[0062] S4. Extract the deep-level attribute features of the input account, and perform semantic mapping with the semantic knowledge graph accordingly, thereby selecting multiple candidate accounts in the semantic knowledge graph that have an intrinsic relationship with the input account that meets the specified requirements.

[0063] The information published by the input account can also undergo the above preprocessing. By mapping with a semantic knowledge graph, we can obtain the deep-level attribute features such as possible topics and opinions in the social text of the input account. Based on the inherent relationship between the basic attribute features and the deep attribute features of the input account, we can filter out potentially similar accounts, i.e., candidate accounts, from the social accounts. The inherent relationship refers to the similarity or compatibility between deep features.

[0064] S5. Perform multi-path attribute combination joint analysis on each candidate account according to the attribute combination scheme, and calculate the similarity between each candidate account and the input account.

[0065] In step S5, the similarity calculation formula is as follows:

[0066]

[0067] In the formula, P represents the similarity between the input account and any candidate account, which is the weighted sum of the similarities from multiple attribute combinations; W g Let be the weight value of any attribute combination g; j is any attribute feature in attribute combination g; w j Let be the weight percentage of the j-th attribute feature in attribute combination g; SIMILAR(·) represents the similarity calculation function, which calculates the similarity value of each attribute feature under each combination, such as the similarity between the names of account a and account b; and Let J represent the j-th attribute feature of the input account and any candidate account, respectively.

[0068] Among them, the weight value W g The value will be dynamically adjusted based on the missing information of that attribute combination. For example, for an attribute combination consisting of name and personal introduction, if the weight of that attribute combination is W... g When both accounts have a name attribute, and one account lacks a personal profile, then W g W will be appropriately reduced g1 Similarly, when both accounts have personal profiles, and one account lacks the name attribute, then W g It will be appropriately reduced to W g2 In this embodiment, W g The calculation formula is:

[0069]

[0070] In the formula, Let g and z represent all the attribute features in the attribute combination g, respectively; Reward(·) represents the weight compensation function, which is used to comprehensively evaluate the weight value of the attribute combination based on the number of missing attributes and the number of hit combinations.

[0071] S6. Arrange all candidate accounts in descending order of similarity into a queue, and select the first specified number of candidate accounts in the queue as similar accounts to the input account.

[0072] First, similarity edges are established with similar accounts in the semantic knowledge graph, and the degree of similarity is labeled. Second, users can filter the Top N accounts based on their own experience, realize the linking of different accounts of users, and use the information differences between different accounts to complete and improve the user profile, so as to have a more comprehensive understanding of the user's social behavior and preferences, thereby better modeling users and achieving accurate prediction of users' social preferences and behaviors.

[0073] Example 2

[0074] This embodiment provides a user identity linking system 10 based on knowledge graph and multi-path attribute fusion, applying the user identity linking method based on knowledge graph and multi-path attribute fusion in embodiment 1.

[0075] Please see Figure 3 The user identity linking system 10 includes: a data acquisition module 101, a graph construction module 102, and a data processing module 103.

[0076] The data acquisition module 101 is used to collect information published by several accounts on different social media platforms to form a dataset;

[0077] The graph construction module 102 is used to construct an account-based semantic knowledge graph based on the dataset. The semantic knowledge graph is used to reflect and store the relationships between various entities. The types of entities include: user entities, account entities, and attribute feature entities. Attribute feature entities include multiple attribute features, each of which is a basic attribute feature or a deep attribute feature.

[0078] The data processing module 103 is used to: first, set up multi-path attribute combinations based on all types of attribute features, and determine the attribute combination scheme from the multi-path attribute combinations according to the information published by the input account; then, extract the deep-level attribute features of the input account, and perform semantic mapping with the semantic knowledge graph, thereby filtering multiple candidate accounts in the semantic knowledge graph that have an intrinsic association with the input account that meets the specified requirements; then, perform multi-path attribute combination joint analysis on each candidate account according to the attribute combination scheme, and calculate the similarity between each candidate account and the input account; finally, arrange all candidate accounts in descending order of similarity into a queue, and select a specified number of candidate accounts at the top of the queue as similar accounts to the input account.

[0079] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A user identity linking method based on knowledge graph and multi-path attribute fusion, characterized in that, include: S1. Collect information published by several accounts on different social media platforms to form a dataset; S2. Construct an account-based semantic knowledge graph based on the dataset; The semantic knowledge graph is used to reflect and store the relationships between various entities; The categories of entities include: user entities, account entities, and attribute feature entities; the attribute feature entities include multiple attribute features, each of which is a basic attribute feature or a deep-level attribute feature. S3. Set up multi-path attribute combinations based on all types of attribute features, and determine the attribute combination scheme from the multi-path attribute combinations according to the information published by the input account; S4. Extract the deep-level attribute features of the input account, and perform semantic mapping with the semantic knowledge graph accordingly, so as to select multiple candidate accounts in the semantic knowledge graph that have an intrinsic relationship with the input account that meets the specified requirements. S5. Perform multi-attribute combination joint analysis on each candidate account according to the attribute combination scheme, and calculate the similarity between each candidate account and the input account. The formula for calculating the similarity is as follows: In the formula, P This indicates the similarity between the input account and any of the candidate accounts; Group For the aforementioned multi-path attribute combination, , Indicates the first q Road attribute combinations q The total number of attribute combinations, where each attribute combination contains at least two attribute features; j For attribute combination Any one of the attribute features in; For the first j Type of attribute features in attribute combination The weighting percentage in; This represents the similarity calculation function; and These represent the first and second entries of the entered account and any alternative account, respectively. j Species characteristics; For any combination of attributes The weight value is calculated using the following formula: ; Representing attribute combinations All attribute features, This indicates the combination of attributes. The total number of attribute features in; This represents the weight compensation function, which is used to comprehensively evaluate the weight value of an attribute combination based on the number of missing attributes and the number of matched combinations. S6. Arrange all candidate accounts in descending order of similarity into a queue, and select a specified number of candidate accounts at the top of the queue as similar accounts to the input account.

2. The user identity linking method based on knowledge graph and multi-path attribute fusion according to claim 1, characterized in that, In step S2, basic attribute features are obtained from the personal information filled in by the user on the social platform; deep attribute features are extracted from the social text formed by the posts and / or comments published by the user on the social platform.

3. The user identity linking method based on knowledge graph and multi-path attribute fusion according to claim 2, characterized in that, In step S2, a fine-tuned large language model is used to perform semantic analysis on the social text, thereby extracting deep-level attribute features.

4. The user identity linking method based on knowledge graph and multi-path attribute fusion according to claim 2, characterized in that, In step S2, the user entity is , The account entity is , ; Attribute feature entity , In the formula, Indicates the first m One user; Indicates the first n One account; Indicates the first p Species characteristics; m , n , p The numbers, in order, represent the total number of users, the total number of accounts, and the total number of attribute features in the semantic knowledge graph.

5. A user identity linking system based on knowledge graph and multi-attribute fusion, characterized in that, Apply the user identity linking method based on knowledge graph and multi-path attribute fusion as described in any one of claims 1 to 4; The user identity linking system includes: The data acquisition module is used to collect information published by several accounts on different social media platforms to form a dataset; The graph construction module is used to construct an account-based semantic knowledge graph based on the dataset. The semantic knowledge graph reflects and stores relationships between various entities. Entity categories include: user entities, account entities, and attribute feature entities. The attribute feature entities include multiple attribute features, each of which is either a basic attribute feature or a deep-level attribute feature. The data processing module is used for: firstly, setting up multi-attribute combinations based on all types of attribute features, and determining the attribute combination scheme from the multi-attribute combinations according to the information published by the input account; secondly, extracting the deep-level attribute features of the input account, and semantically mapping them with the semantic knowledge graph, thereby filtering multiple candidate accounts in the semantic knowledge graph that meet the specified requirements of intrinsic association with the input account; subsequently, performing multi-attribute combination joint analysis on each candidate account according to the attribute combination scheme, and calculating the similarity between each candidate account and the input account; finally, arranging all candidate accounts in descending order of similarity into a queue, and selecting a specified number of candidate accounts at the top of the queue as similar accounts to the input account.

Images