User identity verification method and device, readable storage medium and program product
By generating a high-dimensional vector matrix and using singular value decomposition and TF-IDF model for two-stage recall, the problems of computational complexity and high false positive rate in user authentication are solved, achieving efficient and accurate user authentication.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE PEOPLES INSURANCE CO (GRP) OF CHINA LTD
- Filing Date
- 2025-12-30
- Publication Date
- 2026-05-08
AI Technical Summary
Existing user authentication schemes are computationally complex and have a high false positive rate on large datasets, making it difficult to effectively identify potential malicious users.
A high-dimensional vector matrix is generated by obtaining the character importance of the target text, and the dimensionality is reduced using a singular value decomposition model. The similarity is calculated by combining the TF-IDF model, and two recalls are performed to verify the user's identity.
It reduces computational complexity, improves verification efficiency, ensures rapid response and accuracy on large datasets, and reduces false positive rates.
Smart Images
Figure CN121996776A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence technology, and in particular to a user authentication method and apparatus, a readable storage medium, and a program product. Background Technology
[0002] In various user-facing business domains, authentication is performed on different users accessing the business system to improve its security. For example, short text matching technology can be used for user authentication, matching the user's identity information with a blacklist database to effectively identify potential malicious users. This helps the business system identify possible misconduct and reduce risks.
[0003] Existing short text matching methods include character-based similarity metrics, such as edit distance; tag-based similarity metrics, such as atomic strings; or speech-based similarity metrics, etc. However, different rules are needed to fully cover data ambiguities in different situations. Furthermore, similarity matching has low accuracy, requires the development of corresponding rules to generate a large number of training samples, and has high requirements for sample data, which increases computational complexity.
[0004] Therefore, traditional user authentication schemes suffer from problems such as computational complexity on large datasets and high false positive rates. Summary of the Invention
[0005] The purpose of this application is to provide a user authentication method and apparatus, a readable storage medium, and a program product to solve the problems of computational complexity and high false alarm rate in existing user authentication blacklist text matching.
[0006] To solve the above-mentioned technical problems, this specification is implemented as follows: Firstly, a user authentication method is provided, including: Obtain target text containing the target user's identity information; Based on the importance of each character in the target text, a high-dimensional vector matrix of the first dimension corresponding to the target text is obtained, and a low-dimensional vector matrix of the second dimension is obtained by reducing the high-dimensional vector matrix. Based on the low-dimensional vector matrix and the high-dimensional vector matrix, a similarity match is performed with the blacklist database to retrieve blacklist texts that exceed a preset similarity threshold. The blacklist database pre-stores multiple blacklist texts corresponding to the first-dimensional high-dimensional vector matrix and the second-dimensional low-dimensional vector matrix. The identity of the target user is verified based on the blacklist text that exceeds the preset similarity threshold.
[0007] Optionally, obtaining the high-dimensional vector matrix corresponding to the first dimension of the target text includes: The target text is split into multiple substrings, each substring consisting of one or more characters of the same length; Calculate the importance of each substring in the target text and the multiple blacklist texts; Based on the importance level and the total number of different substrings corresponding to the multiple blacklist texts, a high-dimensional vector matrix of the first dimension corresponding to the target text is obtained, where the first dimension corresponds to the total number of different substrings.
[0008] Optionally, the target text is split into multiple substrings, including: The target text is split into multiple substrings containing characters of the predetermined length by performing a sliding window operation on multiple characters of the predetermined length using an N-gram model.
[0009] Optionally, calculating the importance of each split substring within the target text and the plurality of blacklisted texts includes: The TF-IDF weights of the target substring with respect to the target text and the multiple blacklisted texts are calculated using the TF-IDF model. The TF-IDF weights are used to determine the importance of the target substring in the target text and the multiple blacklist texts.
[0010] Optionally, obtaining the second dimension of the reduced high-dimensional vector matrix into a low-dimensional vector matrix includes: The high-dimensional vector matrix is reduced in dimensionality using a singular value decomposition model to obtain a low-dimensional vector matrix with the second dimension.
[0011] Optionally, a similarity match is performed between the low-dimensional vector matrix and the high-dimensional vector matrix and a blacklist database to retrieve blacklist texts that exceed a preset similarity threshold, including: The low-dimensional vector matrix corresponding to the target text is matched with the low-dimensional vector matrices of the second dimension corresponding to multiple blacklist texts pre-stored in the blacklist database to retrieve the top M most similar blacklist texts. The high-dimensional vector matrix corresponding to the target text is matched with the high-dimensional vector matrix of the first dimension corresponding to the first M most similar blacklist texts pre-stored in the blacklist database to retrieve the first N most similar blacklist texts that exceed the preset similarity threshold, where N is less than M. The top N most similar blacklist texts are identified as the blacklist texts retrieved corresponding to the target text.
[0012] Optionally, before retrieving blacklisted texts that exceed a preset similarity threshold, the process also includes: Based on the importance of each character in the corresponding blacklist text, a high-dimensional vector matrix of the first dimension and a low-dimensional vector matrix of the second dimension corresponding to each blacklist text are obtained.
[0013] In a second aspect, a user authentication device is provided, including a processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions, when executed by the processor, implementing the steps of the method described in the first aspect.
[0014] Thirdly, a readable storage medium is provided that stores a program or instructions which, when executed by a processor, implement the steps of the method described in the first aspect.
[0015] Fourthly, a computer program product is provided, comprising a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform the steps of the method described in the first aspect.
[0016] In this embodiment, target text carrying the identity information of the target user is obtained; based on the importance of each character in the target text, a high-dimensional vector matrix corresponding to the first dimension and a low-dimensional vector matrix of the second dimension are obtained; the low-dimensional vector matrix and the high-dimensional vector matrix are matched with a blacklist database to retrieve blacklist texts that exceed a preset similarity threshold. The blacklist database pre-stores multiple high-dimensional vector matrices and low-dimensional vector matrices corresponding to the first dimension and the second dimension of the blacklist texts; the identity of the target user is verified based on the blacklist texts that exceed the preset similarity threshold. Therefore, a vector matrix similar to the blacklist text can be generated based on the importance of each character in the text, performing two retrievals in the low-dimensional and high-dimensional spaces. This method is computationally efficient and simple to implement. Low-dimensional space computation ensures a fast response speed for real-time queries, while high-dimensional space computation ensures the accuracy of similarity query results. This matching method can significantly reduce the problems of computational complexity and high false positive rates on large datasets. Attached Figure Description
[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating the user authentication method according to an embodiment of this application.
[0018] Figure 2 This is a flowchart illustrating a specific example of a user authentication method according to an embodiment of this application.
[0019] Figure 3 This is a structural block diagram of a user authentication device according to an embodiment of this application. Detailed Implementation
[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. The drawing numbers in this application are only used to distinguish the various steps in the solution and are not used to limit the execution order of the various steps. The specific execution order is subject to the description in the specification.
[0021] To address the problems existing in the prior art, embodiments of this application provide a user authentication method, such as... Figure 1 As shown, it includes the following steps: Step 102: Obtain the target text carrying the target user's identity information; Step 104: Based on the importance of each character in the target text, obtain the high-dimensional vector matrix of the first dimension corresponding to the target text, and the low-dimensional vector matrix of the second dimension after the high-dimensional vector matrix is reduced. Step 106: Based on the low-dimensional vector matrix and the high-dimensional vector matrix, perform similarity matching with the blacklist database to retrieve blacklist texts that exceed a preset similarity threshold. The blacklist database pre-stores multiple blacklist texts corresponding to the first-dimensional high-dimensional vector matrix and the second-dimensional low-dimensional vector matrix. Step 108: Verify the identity of the target user based on the blacklist text that exceeds the preset similarity threshold.
[0022] The target user is a prospective insured user accessing a target application scenario, such as an insurance business system. The target user's identity information includes their insurance application details, such as name and address, and is recorded in text format. To improve the security of the business system, after the target user accesses the system and provides their identity information, the text containing that information can be retrieved for subsequent verification.
[0023] The target text carrying the target user's identity information may be Chinese text and / or English text. Each character in the target text corresponds to a Chinese character or an English word. Different characters have different levels of importance in the target text. In this step, the importance of multiple characters in the target text can be calculated, a high-dimensional vector matrix corresponding to the first dimension of the target text can be established, and a low-dimensional vector matrix with reduced dimensions can be established based on the high-dimensional vector matrix.
[0024] Clearly, the second dimension is inferior to the first dimension, which is related to the number of characters in the target text.
[0025] Based on the solution provided in the above embodiments, optionally, in step 104 above, obtaining a high-dimensional vector matrix corresponding to the first dimension of the target text based on the importance of each character included in the target text in the target text includes: splitting the target text into multiple substrings, each substring including one or more characters of the same length; calculating the importance of each split substring in the target text and the multiple blacklist texts; and obtaining a high-dimensional vector matrix corresponding to the first dimension of the target text based on the importance and the total number of different substrings corresponding to the multiple blacklist texts, wherein the first dimension corresponds to the total number of different substrings.
[0026] In this embodiment, the dimension of the high-dimensional vector matrix corresponding to the target text is the same as the number of substrings obtained by splitting the target text. For example, if the target text is a user's name containing 3 Chinese characters, it can be split into 3 individual Chinese characters (corresponding to 3 substrings), or into 2 words containing 2 Chinese characters (corresponding to 2 substrings). The splitting criterion is that each split substring contains characters of the same length, either 1 Chinese character or 2 Chinese characters.
[0027] Then, determine the importance of each of the split substrings in the target text. If the number of substrings is 2, then calculate 2 vector values representing the importance.
[0028] Similarly, multiple blacklist texts pre-stored in the blacklist database can be split into corresponding substrings. Different blacklist texts may be split into one or more identical substrings. The total number of different substrings obtained from splitting multiple blacklist texts is counted. For example, if a blacklist text contains a first substring after splitting, and other blacklist texts also contain the same first substring after splitting, then the number of multiple identical first substrings is counted as 1. If there are n different substrings, then the total number of different substrings corresponding to the splitting of multiple blacklist texts is n.
[0029] Return to the target text. The vector values of the two split substrings are combined with the total number n of different substrings corresponding to the multiple blacklist texts pre-stored in the blacklist database to obtain a high-dimensional vector matrix with a corresponding dimension of n, that is, a high-dimensional vector matrix of 1*n, where 1 represents one target text and n represents the total number of n different substrings of the blacklist text.
[0030] In the 1*n high-dimensional vector matrix, the substrings that are the same as the two substrings split from the target text have vector values corresponding to represent the degree of importance, and the vector values corresponding to the substrings that are different from the two substrings split from the target text are 0. That is, the 1*n high-dimensional vector matrix is a sparse matrix, and most of the vector values are 0.
[0031] Specifically, in one embodiment, splitting the target text into multiple substrings includes: performing a sliding window operation of a predetermined length on the multiple characters included in the target text through an N-gram model, and splitting the target text into multiple substrings including characters of the preset length.
[0032] N-gram is an algorithm based on a statistical language model. Its basic idea is to perform a sliding window operation of length N on the content in the text according to characters, forming a sequence of substrings of length N. For example, when N = 2, the text "Hello World" will be split into three substrings: "Hello", "elloW", and "World".
[0033] In this embodiment, by performing a sliding window operation of a predetermined length on the multiple characters included in the target text, for application scenarios where the target text is a Chinese text and an English text, the value of N can be different. For example, N can be 2 in Chinese texts and N can be 3 in English texts. Thus, the target text carrying the identity information of the target user can be split into multiple substrings including characters of the preset length.
[0034] Furthermore, calculating the degree of importance of each split substring in the target text and the multiple blacklist texts includes: calculating the TF-IDF weights of the target substring for the target text and the multiple blacklist texts through the TF-IDF model; determining the degree of importance of the target substring in the target text and the multiple blacklist texts based on the TF-IDF weights.
[0035] The Term Frequency–Inverse Document Frequency (TF-IDF) model is used to evaluate the importance of a word to a document within a document set or corpus. It is a weighted technique used in information retrieval and data mining. The importance of a word increases proportionally to the number of times it appears in a document, and decreases inversely proportionally to its frequency in the corpus.
[0036] In this embodiment, the method can be used to calculate the importance of each substring included in the target file to the target file and multiple blacklisted texts. The TF-IDF weights of each split substring to the target text and multiple blacklisted texts are calculated using a TF-IDF model, and the TF-IDF weights are determined as vector values representing the importance of the target substring in the target text and multiple blacklisted texts.
[0037] As mentioned above, the target text can be represented as a 1*n high-dimensional vector matrix, where each dimension of the vector corresponds to a substring, and its value is the TF-IDF weight of that substring. Assuming the total number of blacklisted texts is m and the total number of distinct substrings is n, the constructed high-dimensional vector matrix of the target text's TF-IDF weights has a dimension of 1*n. This matrix is highly sparse, effectively storing the weighted feature representation of the target text.
[0038] Specifically, the TF-IDF weights are determined based on the product of term frequency (TF) and inverse document frequency (IDF).
[0039]
[0040]
[0041]
[0042] in, This represents the word frequency of a substring in the target text relative to the entire target text. This represents the inverse text frequency index of the substring in the target text relative to multiple blacklisted texts. This indicates the TF-IDF weight of the substring in the target text relative to the target text.
[0043] For multiple substrings included in the target text, their corresponding TF-IDF weights can be calculated using the same method described above. This yields a high-dimensional vector matrix, for example, 1*n, representing the importance of each substring within the target text.
[0044] In one specific embodiment, obtaining the second dimension of the low-dimensional vector matrix after dimensionality reduction of the high-dimensional vector matrix includes: reducing the dimensionality of the high-dimensional vector matrix using a singular value decomposition model to obtain the second dimension of the low-dimensional vector matrix.
[0045] In this implementation, the dimensionality of a high-dimensional vector matrix that is a sparse matrix is reduced by using the Singular Value Decomposition (SVD) model.
[0046] The high-dimensional vector matrix of TF-IDF weights corresponding to the target text has a large dimensionality space, which takes a long time to calculate. Here, we use the singular value decomposition model to reduce the dimensionality of the 1*n TF-IDF weight matrix (assuming that the second dimension after dimensionality reduction is k) to reduce the computational requirements.
[0047] Truncated Singular Value Decomposition (SVD) reduces a high-dimensional 1*n vector matrix to a low-dimensional 1*k vector matrix. SVD decomposition can be represented as follows:
[0048] Where U is an orthogonal matrix, and its columns are called left singular vectors; Σ is the transpose of V, and is also an orthogonal matrix, whose columns are called right singular vectors; Σ is a diagonal matrix, whose diagonal elements are called singular values. When m=1, it corresponds to a high-dimensional vector matrix of dimension n corresponding to the target text.
[0049] To achieve dimensionality reduction, only the k largest singular values and their corresponding singular vectors are retained. The vectors in this low-dimensional vector matrix are a linear transformation of the vectors in the original high-dimensional vector matrix, condensing the vectors of the most discriminative substrings. After truncation using the singular value decomposition model, the dimensionality is reduced to:
[0050] When m=1, it corresponds to a low-dimensional vector matrix of dimension k corresponding to the target text.
[0051] Based on the solution provided in the above embodiments, optionally, in step 106 above, performing similarity matching between the low-dimensional vector matrix and the high-dimensional vector matrix and the blacklist database to retrieve blacklist texts exceeding a preset similarity threshold includes: performing similarity matching between the low-dimensional vector matrix corresponding to the target text and the low-dimensional vector matrices of the second dimension corresponding to multiple blacklist texts pre-stored in the blacklist database to retrieve the top M most similar blacklist texts; performing similarity matching between the high-dimensional vector matrix corresponding to the target text and the high-dimensional vector matrices of the first dimension corresponding to the top M most similar blacklist texts pre-stored in the blacklist database to retrieve the top N most similar blacklist texts exceeding the preset similarity threshold, where N is less than M; and determining the top N most similar blacklist texts as the blacklist texts retrieved corresponding to the target text.
[0052] In this embodiment, firstly, in the low-dimensional space corresponding to the low-dimensional vector matrix, the target text and multiple blacklist texts pre-stored in the blacklist database are matched for similarity to perform the first recall and retrieve the top M most similar blacklist texts.
[0053] Then, in the high-dimensional space of the high-dimensional vector matrix corresponding to the top M most similar blacklist texts, the target text and the top M most similar blacklist texts are matched for similarity to perform a second recall, and the top N most similar blacklist texts that exceed the preset similarity threshold are retrieved from the top M most similar blacklist texts, where N is less than M.
[0054] Finally, the top N most similar blacklisted texts obtained from the two recalls are used as the blacklisted texts retrieved for the target text.
[0055] In the above embodiments, similarity matching calculation is performed based on the low-dimensional vector matrix of the target file and the low-dimensional vector matrices of the same dimension corresponding to multiple blacklist texts, and similarity matching calculation is performed based on the high-dimensional vector matrix of the target file and the high-dimensional vector matrices of the same dimension corresponding to the top M most similar blacklist texts.
[0056] In one specific embodiment, the similarity matching calculation can sample a suitable similarity algorithm, such as the IndexFlatL2 algorithm based on Euclidean distance, or the IndexFlatIP algorithm based on cosine similarity.
[0057] The formula for the IndexFlatL2 algorithm is as follows:
[0058] Where q represents the target text and d represents the blacklist text. The distance between the target text and the corresponding blacklist text is represented by k, where k represents the total number of dimensions of the low-dimensional vector matrix, and i represents the i-th dimension.
[0059] The formula for the IndexFlatIP algorithm is as follows:
[0060] Among them, q and d have been standardized. This represents the cosine similarity between the target text and the corresponding blacklist text.
[0061] The high-dimensional vector matrix and low-dimensional vector matrix corresponding to the blacklist text in the blacklist database are determined using the same method as the target text mentioned above.
[0062] Optionally, before retrieving blacklisted texts that exceed a preset similarity threshold, the method further includes: based on the importance of each character included in the multiple blacklisted texts in the corresponding blacklisted texts, obtaining a high-dimensional vector matrix of the first dimension corresponding to each blacklisted text, and a low-dimensional vector matrix of the second dimension of the high-dimensional vector matrix after dimensionality reduction.
[0063] Based on the importance of each character in each blacklist text, we obtain a high-dimensional vector matrix (first dimension) and a low-dimensional vector matrix (second dimension) corresponding to each blacklist text. For example, if we treat m blacklist texts in the blacklist database as a whole to determine the high-dimensional vector matrix, the corresponding high-dimensional vector matrix is m*n, and the corresponding low-dimensional vector matrix is m*k.
[0064] Therefore, calculating similarity in low-dimensional space ensures a faster response speed for real-time queries, while calculating similarity in high-dimensional space ensures the accuracy of query results.
[0065] After retrieving the blacklisted texts through the two recall processes described above, only the top N blacklisted texts with similarity scores exceeding a preset similarity threshold can be returned and used to verify the target user's identity. Based on the content of the top N blacklisted texts, it is possible to effectively identify whether the target user is a potential malicious customer, thereby identifying possible misconduct and reducing risk.
[0066] The following is combined with Figure 2 The flowchart of a specific example of a user authentication method in this application is described.
[0067] like Figure 2 As shown, it includes the following steps 202 to 212.
[0068] Step 202, Text Preprocessing: Data cleaning, including special character replacement, Chinese and English character recognition, and abnormal string checking. This text preprocessing is applicable to target text carrying the target user's identity information requested during real-time queries, and also to blacklist text included in the blacklist database.
[0069] Step 204: Construct a sparse matrix (i.e., a high-dimensional vector matrix) based on TF-IDF weights: For Chinese and English texts (including target text and blacklist text), select N-grams of appropriate character lengths to split into multiple substrings, and calculate the TF-IDF value (i.e., weight) of the substrings to construct sparse matrix representations of Chinese and English texts.
[0070] Step 206, SVD dimensionality reduction: Singular value decomposition is used to reduce the dimensionality of the high-dimensional sparse matrix, and the low-dimensional vector matrix is stored in a blacklist database, such as the Faiss database.
[0071] Step 208, First Recall: Using the Euclidean distance index of the Faiss library, recall the top M blacklisted texts that are most similar to the target text in the low-dimensional space.
[0072] During real-time queries, for the requested target text, after processing in steps 202 to 206, a similarity search using the Faiss library is performed to recall and rank the top M matching blacklist texts in the low-dimensional space. Specifically, the requested target text is transformed into a k-dimensional low-dimensional vector matrix using the same TF-IDF and SVD models; then, the Faiss library searches among all m low-dimensional vector matrices for the top M blacklist entries most similar to the low-dimensional vector matrix of the queried target text, and returns the index or number of the corresponding blacklist entry.
[0073] Step 210, Second Recall: Based on the top M blacklists from the first recall, recall the top M blacklists most similar to the target text in the high-dimensional space.
[0074] For the first M matching blacklist texts that have been recalled, they are re-sorted. In high-dimensional space, calculations are performed based on the high-dimensional vector matrix before dimensionality reduction to recall the first N matching texts. That is, from the high-dimensional vector matrix prepared in step 204, a smaller candidate high-dimensional vector matrix M*n is formed based on the first M matching blacklist texts recalled. At the same time, the target text is converted into a corresponding 1*n high-dimensional vector matrix. Finally, the similarity (e.g., cosine similarity) between this 1*n high-dimensional vector matrix and each blacklist text in the candidate high-dimensional vector matrix M*n is calculated, and the texts are sorted according to the similarity scores.
[0075] Step 212, Threshold setting and result return: Set a similarity score threshold to control the return of only the N most matching blacklist texts with similarity scores exceeding the threshold, thereby achieving a balance between precision and recall and controlling the accuracy of the returned blacklist texts.
[0076] In this embodiment, target text carrying the identity information of the target user is obtained; based on the importance of each character in the target text, a high-dimensional vector matrix corresponding to the first dimension and a low-dimensional vector matrix of the second dimension are obtained; the low-dimensional vector matrix and the high-dimensional vector matrix are matched with a blacklist database to retrieve blacklist texts that exceed a preset similarity threshold. The blacklist database pre-stores multiple high-dimensional vector matrices and low-dimensional vector matrices corresponding to the first dimension and the second dimension of the blacklist texts; the identity of the target user is verified based on the blacklist texts that exceed the preset similarity threshold. Therefore, a vector matrix similar to the blacklist text can be generated based on the importance of each character in the text, performing two retrievals in the low-dimensional and high-dimensional spaces. This method is computationally efficient and simple to implement. Low-dimensional space computation ensures a fast response speed for real-time queries, while high-dimensional space computation ensures the accuracy of similarity query results. This matching method can significantly reduce the problems of computational complexity and high false positive rates on large datasets.
[0077] Optionally, such as Figure 3 As shown in the figure, this application embodiment also provides a user authentication device 2000, including a processor 2400 and a memory 2200. The memory 2200 stores a program or instructions that can run on the processor 2400. When the program or instructions are executed by the processor 2400, they implement the various steps of the above-described user authentication method embodiment and can achieve the same technical effect. To avoid repetition, they will not be described again here.
[0078] This application also provides a readable storage medium storing a program or instructions. When executed by a processor, the program or instructions implement the various processes of any of the above-described user authentication method embodiments and achieve the same technical effect. To avoid repetition, further details are omitted here. The readable storage medium includes computer-readable storage media, such as read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0079] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. The computer program is operable to enable a computer to execute various processes of any of the above-described user authentication method embodiments and achieve the same technical effect. To avoid repetition, it will not be described again here.
[0080] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0081] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.
[0082] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.
Claims
1. A user authentication method, characterized in that, include: Obtain target text containing the target user's identity information; Based on the importance of each character in the target text, a high-dimensional vector matrix of the first dimension corresponding to the target text is obtained, and a low-dimensional vector matrix of the second dimension is obtained by reducing the high-dimensional vector matrix. Based on the low-dimensional vector matrix and the high-dimensional vector matrix, a similarity match is performed with the blacklist database to retrieve blacklist texts that exceed a preset similarity threshold. The blacklist database pre-stores multiple blacklist texts corresponding to the first-dimensional high-dimensional vector matrix and the second-dimensional low-dimensional vector matrix. The identity of the target user is verified based on the blacklist text that exceeds the preset similarity threshold.
2. The method according to claim 1, characterized in that, Obtaining the high-dimensional vector matrix corresponding to the first dimension of the target text includes: The target text is split into multiple substrings, each substring consisting of one or more characters of the same length; Calculate the importance of each substring in the target text and the multiple blacklist texts; Based on the importance level and the total number of different substrings corresponding to the multiple blacklist texts, a high-dimensional vector matrix of the first dimension corresponding to the target text is obtained, where the first dimension corresponds to the total number of different substrings.
3. The method according to claim 2, characterized in that, The target text is split into multiple substrings, including: The target text is split into multiple substrings, each containing characters of the predetermined length, by performing a sliding window operation of a predetermined length on the multiple characters included in the target text using an N-gram model.
4. The method according to claim 2, characterized in that, Calculating the importance of each segmented substring within the target text and the multiple blacklisted texts includes: The TF-IDF weights of the target substring with respect to the target text and the multiple blacklisted texts are calculated using the TF-IDF model. The TF-IDF weights are used to determine the importance of the target substring in the target text and the multiple blacklist texts.
5. The method according to claim 1, characterized in that, The process of obtaining the second dimension of the reduced-dimensional vector matrix from the high-dimensional vector matrix includes: The high-dimensional vector matrix is reduced in dimensionality using a singular value decomposition model to obtain a low-dimensional vector matrix with the second dimension.
6. The method according to claim 1, characterized in that, Based on the low-dimensional vector matrix and the high-dimensional vector matrix, a similarity match is performed with the blacklist database to retrieve blacklist texts that exceed a preset similarity threshold, including: The low-dimensional vector matrix corresponding to the target text is matched with the low-dimensional vector matrices of the second dimension corresponding to multiple blacklist texts pre-stored in the blacklist database to retrieve the top M most similar blacklist texts. The high-dimensional vector matrix corresponding to the target text is matched with the high-dimensional vector matrix of the first dimension corresponding to the first M most similar blacklist texts pre-stored in the blacklist database to retrieve the first N most similar blacklist texts that exceed the preset similarity threshold, where N is less than M. The top N most similar blacklist texts are identified as the blacklist texts retrieved corresponding to the target text.
7. The method according to any one of claims 1 to 6, characterized in that, Before retrieving blacklisted texts that exceed a preset similarity threshold, the process also includes: Based on the importance of each character in the corresponding blacklist text, a high-dimensional vector matrix of the first dimension and a low-dimensional vector matrix of the second dimension corresponding to each blacklist text are obtained.
8. A user authentication device, characterized in that, It includes a processor and a memory, the memory storing a program or instructions that can run on the processor, the program or instructions being executed by the processor to implement the steps of the method as described in any one of claims 1-7.
9. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1-7.
10. A computer program product, characterized in that, The computer program product includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform the steps of the method as described in any one of claims 1-7.