An archive management system based on cloud archives

Abstract

The present invention relates to the field of archive management technology, and specifically discloses an archive management system based on a cloud archive library, comprising: a collection and processing module: generating a pending vector based on an access keyword sequence of an incoming user; obtaining a comparison vector of a target user; obtaining a cosine similarity A between the pending vector and the comparison vector, and setting a cosine similarity threshold Ays; a first partitioning module: when the cosine similarity A>Ays, treating the corresponding target user as a merged user, adding supplementary content B to the merged user's data partition to obtain a data partition X, and treating data partition X as a common data partition for the merged user and the incoming user; and a second partitioning module: when the cosine similarity A≤Ays, obtaining a target keyword, and creating a separate data partition for the incoming user based on the target keyword. The present invention can dynamically construct and merge partitions, saving space and operation and maintenance resources while maintaining fine-grained permission isolation and efficient and secure access.

Description

Technical Field

[0001] The present invention relates to the technical field of archive management, and in particular to an archive management system based on a cloud archive library. Background Art

[0002] Cloud archives are a digital information management model that migrates traditional paper or localized electronic archives to a cloud computing architecture. Leveraging distributed storage, elastic computing, and multiple backup technologies, they centrally host massive amounts of archival data in the cloud, enabling anytime, anywhere access and collaboration.

[0003] Data partitioning is a management strategy within a cloud archive that isolates and divides cloud storage and computing resources into logical units based on factors such as department, business domain, or user identity. By creating a data partition for each user, physical and logical isolation of archival data, refined permission control, and multiple independent backups can be achieved within the same distributed storage and elastic computing architecture. However, when a large number of people access the archive, creating more data partitions will result in increased storage space and system resource usage. Summary of the Invention

[0004] The purpose of the present invention is to provide a cloud-based archive management system to solve the above technical problems.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A cloud archive management system, comprising:

[0007] Collection and processing module: When a user enters the cloud archive, they are marked as entering users, and a pending vector is generated based on the access keyword sequence of the entering user; entering users who have opened data partitions and are online are marked as target users, and the comparison vector of the target user is obtained;

[0008] Obtaining the cosine similarity A between the pending vector and the comparison vector, and setting a cosine similarity threshold Ays;

[0009] First partitioning module: When the cosine similarity A>Ays, the corresponding target user is taken as the merged user, and the supplementary content B is added to the data partition of the merged user to obtain a data partition X. The data partition X is used as the common data partition of the merged user and the entering user. The supplementary content B∈B1 and B∉B2, where B1 and B2 represent the content allowed to be accessed by the entering user and the merged user, respectively;

[0010] Second partitioning module: when the cosine similarity A≤Ays, a target keyword is obtained based on the browsing history of the entering user in the cloud archive, and a separate data partition is opened for the entering user based on the target keyword.

[0011] As a further solution of the present invention, obtaining the undetermined vector and the comparison vector includes:

[0012] For the incoming user, convert a single keyword in the corresponding access keyword sequence into a word vector, perform TF-IDF weighted averaging on all the word vectors to obtain a target vector, which is recorded as the undetermined vector;

[0013] Obtain the target vector of the target user and record it as a comparison vector.

[0014] As a further solution of the present invention, opening a separate data partition for the incoming user includes:

[0015] Obtaining the browsing history of the entering user within a preset monitoring period, marking a single file in the browsing history as a target file, and taking the browsed portion of the single target file as the target portion;

[0016] Based on the target part, the target keywords are determined, and the archives in the cloud archive library whose types and quantities of the target keywords exceed the preset values ​​are marked as partition archives. A data partition Y is opened, and the partition archives are stored in the data partition Y. The data partition Y is used as the data partition for the user.

[0017] As a further solution of the present invention: determining the target keyword includes:

[0018] Extracting keywords from the target portion and recording them as pending words, grouping the pending words, wherein different groups contain different numbers and / or types of pending words;

[0019] Determine the number n of undetermined words in target part i contained in group j, and calculate the degree of fit k between group j and target part i ij =n / N j , N j Indicates the number of undetermined words in the group j, and calculates the total fit K of group j j ;

[0020] Generate the total fit set K jh = (K1, K2, ..., K m ), m represents the total number of groups, and the maximum total fit Kmax=max(K jh ) The pending words in the corresponding group are used as target keywords.

[0021] As a further solution of the present invention: when the data partition corresponding to the merged user corresponds to the remaining target users, the following steps are performed:

[0022] The remaining target users corresponding to the data partition corresponding to the merged user are marked as pending users. When the cosine similarity between the comparison vector of any pending user and the pending vector is less than the cosine similarity threshold, the merged user is no longer considered as a merged user.

[0023] As a further solution of the present invention: when two or more cosine similarities are greater than the cosine similarity threshold, a prompt message is sent for reporting.

[0024] As a further solution of the present invention: when the target user does not exist, the steps in the second partitioning module are executed to open a data partition for the incoming user.

[0025] As a further solution of the present invention: when the registration time of the entering user is less than the preset duration, the subsequent steps are not executed, and a prompt message is sent to the preset administrator, who manually opens the data partition of the entering user.

[0026] The beneficial effects of the present invention are as follows:

[0027] 1) This invention calculates the vector cosine similarity between incoming users and online users. When a threshold condition is met, the two are merged into the same data partition and only the differentiated content is added, avoiding the need to repeatedly create independent partitions for each new user. This dynamic merging mechanism enables the number of partitions to converge adaptively with similarity requirements, significantly reducing disk partitioning and metadata overhead, fundamentally alleviating the risk of unnecessary storage space and system resource usage, and keeping the archive structure concise and controllable.

[0028] 2) When the similarity is not enough to trigger a merge, the target keywords are automatically extracted based on the user's actual browsing fragments during the monitoring period, and then relevant archives are screened from the entire library according to the keyword density to instantly create an exclusive partition for the user. This content-driven partition generation method can accurately fit the user's real search interests; at the same time, the correspondence between keywords and archives can be continuously updated with user behavior, and the system automatically adjusts the partition content without the need for frequent manual migration or reconstruction, realizing dynamic alignment of archive resources and business needs, maintaining efficient, refined and sustainable management granularity, significantly shortening the search path, and improving response flexibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present invention will be further described below with reference to the accompanying drawings.

[0030] Figure 1 It is a flow chart of an archive management system based on a cloud archive library of the present invention. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0032] See also Figure 1 As shown, the present invention is a file management system based on a cloud archive library, the present invention is a file management system based on a cloud archive library, comprising:

[0033] Collection and processing module: When a user enters the cloud archive, they are marked as entering users, and a pending vector is generated based on the access keyword sequence of the entering user; entering users who have opened data partitions and are online are marked as target users, and the comparison vector of the target user is obtained;

[0034] In a preferred embodiment of the present invention, obtaining the undetermined vector and the comparison vector includes:

[0035] For the incoming user, convert a single keyword in the corresponding access keyword sequence into a word vector, perform TF-IDF weighted averaging on all the word vectors to obtain a target vector, which is recorded as the undetermined vector;

[0036] Obtaining a target vector of the target user, recorded as a comparison vector;

[0037] It should be noted that when a user is detected logging into the cloud archive, it is first marked as an "entering user" in the session context, and the access keyword sequence is extracted from its most recent search, click or filter operation, such as "contract", "lease", "2023", and then the pre-trained Chinese word vector model (such as Word2Vec or BERTTokenEmbedding) is called to map each keyword to a word vector of the same dimension; the system synchronously queries the word frequency and document frequency of the entire library corpus, and after obtaining the corresponding TF values ​​for the keywords "contract", "lease", and "2023", , multiplied by its IDF value to obtain the weight, and then the three groups of vectors are weighted one by one according to the "word vector × weight" method and summed up. Finally, divided by the sum of the weights, the weighted average vector representing the user's interest intention is obtained, and the vector is stored in the cache and marked as "pending vector". At the same time, the system traverses other users who have created data partitions and are online, and directly loads the target vectors generated and persisted when they enter into the memory, marking them as "comparison vectors". After completing the preparation of the pending vector and multiple comparison vectors, the cosine similarity calculation and partition merging or independent decision-making can be performed subsequently.

[0038] By first mapping each access keyword sequence into a word vector and then performing a weighted average using TF-IDF weights, we can highlight the key information in each user's search intent while maintaining a fixed vector dimension. This reduces the noise introduced by stop words or generalized words, ensuring that the resulting undetermined and comparison vectors possess both holistic semantic expression and word importance. This allows for a more accurate assessment of the true proximity between different users' access needs when subsequently calculating cosine similarity. This approach requires minimal computation and is easily updated online in real time, enabling the system to quickly determine similarity and make partition merging decisions in multi-person concurrent scenarios.

[0039] Obtaining the cosine similarity A between the pending vector and the comparison vector, and setting a cosine similarity threshold Ays;

[0040] First partitioning module: When the cosine similarity A>Ays, the corresponding target user is taken as the merged user, and the supplementary content B is added to the data partition of the merged user to obtain a data partition X. The data partition X is used as the common data partition of the merged user and the entering user. The supplementary content B∈B1 and B∉B2, where B1 and B2 represent the content allowed to be accessed by the entering user and the merged user, respectively;

[0041] For example, when the system detects that the cosine similarity between the incoming user and a target user exceeds a threshold Ays, the target user is identified as a merged user. For example, if the incoming user U1 has a high similarity with the online user U2, the system first searches U1's accessed content set B1 and U2's existing content set B2, and calculates the difference set B = the part that exists in B1 but not in B2. Assuming that B1 contains files a, b, and c and B2 only contains a and b, then B is c. The system then creates or mounts file c in the physical partition storing U2's data. Soft links, hard links, or block-level references can be used instead of full copies to quickly generate the expanded partition X.

[0042] This approach avoids duplicating physical space for each new user with similar needs. By appending only differentiated content rather than replicating the entire file system, data redundancy and disk fragmentation are reduced, keeping the total number of partitions and the size of each partition under control. The centralized file storage enabled by partition sharing simplifies operations such as directory indexing, backup, and recovery. The system eliminates the need to frequently create, mount, and scan large numbers of partitions in high-concurrency scenarios, helping the overall solution achieve comprehensive optimization of storage space, I / O scheduling, and management processes.

[0043] Second partitioning module: when the cosine similarity A≤Ays, obtaining target keywords based on the browsing history of the entering user in the cloud archive, and opening a separate data partition for the entering user based on the target keywords;

[0044] In another preferred embodiment of the present invention, opening a separate data partition for the incoming user includes:

[0045] Obtaining the browsing history of the entering user within a preset monitoring period, marking a single file in the browsing history as a target file, and taking the browsed portion of the single target file as the target portion;

[0046] Determining target keywords based on the target portion, marking archives in the cloud archive library whose types and quantities of the target keywords exceed a preset value as partition archives, creating a data partition Y, storing the partition archives in the data partition Y, and using the data partition Y as the data partition for the incoming user;

[0047] For example, the platform first continuously captures every page access record of the user within a set monitoring period. For example, if a user opens three types of documents, namely "project contract", "budget approval form" and "progress weekly report" multiple times within three days, the full archive entry containing them is marked as the target archive, and the text, attachments, pictures, etc. that the user actually scrolls or searches to locate are recorded as the target part; the platform then performs Chinese word segmentation and named entity recognition on these target parts, calculates the word frequency and position weight after filtering out stop words, and extracts keywords that reflect the user's focus, such as "contract amount", "payment node" and "progress milestone"; then searches the entire database for all archives that contain these keywords and whose frequency exceeds the system threshold, and classifies the qualified entries into partition archives, such as retrieving contracts, supplementary agreements, payment vouchers and corresponding weekly reports of the same series; then creates a new data partition Y, stores the retrieved partition archives into the partition as symbolic links or physical copies, and writes an independent partition mapping pointing to Y in the access control table for the user entering, so that the user can directly land in a dedicated archive collection that is highly consistent with his or her interests after subsequent login;

[0048] It is important to note that using users' actual browsing behavior to drive keyword extraction and archive aggregation can ensure that partition content accurately reflects users' current business tasks, avoiding access redundancy or gaps caused by static allocation based on organizational roles or initial application lists. At the same time, using keyword coverage and occurrence types as thresholds to filter archives ensures that partitions are of moderate size without excessive loss of key information, shortening subsequent search paths and improving cache and index hit rates. Partitions are automatically updated as user interests change, eliminating the need for administrators to frequently manually migrate them. This reduces both the operational and maintenance burden and the risk of storage fragmentation, providing efficient and sustainable support for on-demand partitioning and dynamic scaling of the entire cloud archive.

[0049] In a preferred embodiment of this embodiment, determining the target keyword includes:

[0050] Extracting keywords from the target portion and recording them as pending words, grouping the pending words, wherein different groups contain different numbers and / or types of pending words;

[0051] Determine the number n of undetermined words in target part i contained in group j, and calculate the degree of fit k between group j and target part i ij =n / N j , N j Indicates the number of undetermined words in the group j, and calculates the total fit K of group j j ;

[0052] Generate the total fit set K jh = (K1, K2, ..., K m ), m represents the total number of groups, and the maximum total fit Kmax=max(K jh ) The undetermined words in the corresponding group are used as target keywords;

[0053] It should be noted that several pending words are first extracted from the target part, such as the four keywords "contract amount", "payment node", "progress milestone", and "acceptance criteria". Then, they are grouped one by one according to mathematical permutations and combinations: the four words are first considered as a complete combination, and then all triple combinations are enumerated, such as "contract amount + payment node + progress milestone". The remaining groups are not repeated. The fit is defined as the ratio of the number of words in the combination to the size of the combination. It is essentially a coverage metric: it not only measures the completeness of a keyword combination in a single browsing segment, but also normalizes combinations of different sizes through the denominator, so that combinations containing two, three, or even all words can be compared on the same scale, thereby avoiding the bias of naturally obtaining higher absolute match counts due to a large number of words. The coverage is accumulated over all target parts to further examine the overall adaptability of the combination to the user's continuous browsing behavior, reflecting its stability and lasting relevance. Finally, the combination with the highest overall fit is selected, which is equivalent to selecting the most frequently mentioned and relatively complete keyword set, thereby maximizing the restoration of user attention topics in a statistical sense.

[0054] When the data partition corresponding to the merged user corresponds to the remaining target users, perform the following steps:

[0055] Mark the remaining target users corresponding to the data partition corresponding to the merged user as pending users. If the cosine similarity between the comparison vector of any of the pending users and the pending vector is less than the cosine similarity threshold, the merged user is no longer considered as a merged user.

[0056] It is understandable that when the merged user already shares the same data partition with other target users, the system will mark all of these users as pending users and then calculate the semantic vector similarity between each pending user and the newly entered user. If the similarity between any pending user and the newly entered user falls below the set threshold, the merge will be immediately canceled, maintaining the existing partition structure. This ensures that all users within the same data partition maintain a high degree of consistency in their access needs, preventing unrelated users from being forcibly aggregated.

[0057] It should be noted that when two or more cosine similarities are greater than the cosine similarity threshold, a prompt message is sent for reporting;

[0058] When the semantic similarity between two or more users and a newly entered user exceeds the preset threshold at the same time, an immediate report is generated, allowing administrators to immediately focus on possible group high-overlap access situations and promptly investigate risk scenarios such as account fraud, data leakage, or abnormal collaboration, thereby preventing the system from automatically merging and forming an excessively large shared scope, which would weaken the isolation strength.

[0059] It should be noted that, when the target user does not exist, the steps in the second partitioning module are executed to open a data partition for the incoming user.

[0060] When there are no online target users in the database for comparison, the system creates an independent partition for the incoming user, ensuring that their legitimate business operations are not delayed. At the same time, it avoids rigidly binding users with low relevance or unknown needs to previous partitions, which generates redundant data. In these two extreme cases, the security closed loop is completed through manual intervention and automatic implementation, respectively. This balances real-time and precision, while ensuring the integrity and sustainability of the cloud archive's overall isolation strategy.

[0061] It is understandable that when the registration time of the user entering the system is less than the preset time, the subsequent steps will not be executed, and a prompt message will be sent to the preset administrator, who will manually open the data partition for the user entering the system.

[0062] When a user has registered for too short a time, there is insufficient data to support their identity credibility, access intention, and subsequent activity. If the automatic partitioning process is entered immediately, the similarity may be misjudged due to the lack of historical keywords, resulting in unauthorized use. Therefore, the system first suspends automatic processing and pushes the information to the administrator for manual review and manual creation of partitions. This can not only verify the user's identity, business background, and the required scope of files, prevent malicious registration and temporary accounts from quickly grabbing sensitive files, but also lay an accurate starting point for subsequent automatic merging or independent partitioning, helping the cloud archive to maintain a reasonable number of partitions and efficient use of storage resources while ensuring security and compliance.

[0063] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A cloud archive management system, characterized in that: include: Collection and processing module: When a user enters the cloud archive, they are marked as entering users, and a pending vector is generated based on the access keyword sequence of the entering user; entering users who have opened data partitions and are online are marked as target users, and the comparison vector of the target user is obtained; Obtaining the cosine similarity A between the pending vector and the comparison vector, and setting a cosine similarity threshold Ays; First partitioning module: When the cosine similarity A>Ays, the corresponding target user is taken as the merged user, and the supplementary content B is added to the data partition of the merged user to obtain a data partition X. The data partition X is used as the common data partition of the merged user and the entering user. The supplementary content B∈B1 and B∉B2, where B1 and B2 represent the content allowed to be accessed by the entering user and the merged user, respectively; Second partitioning module: when the cosine similarity A≤Ays, obtaining target keywords based on the browsing history of the entering user in the cloud archive, and opening a separate data partition for the entering user based on the target keywords; Opening a separate data partition for the incoming user includes: Obtaining the browsing history of the entering user within a preset monitoring period, marking a single file in the browsing history as a target file, and taking the browsed portion of the single target file as the target portion; Determining target keywords based on the target portion, marking archives in the cloud archive library whose types and quantities of the target keywords exceed a preset value as partition archives, creating a data partition Y, storing the partition archives in the data partition Y, and using the data partition Y as the data partition for the incoming user; Target keywords include: Extracting keywords from the target portion and recording them as pending words, grouping the pending words, wherein different groups contain different numbers and / or types of pending words; Determine the number n of undetermined words in target part i contained in group j, and calculate the degree of fit k between group j and target part i ij =n / N j , N j Indicates the number of undetermined words in the group j, and calculates the total fit K of group j j ; Generate the total fit set K jh = (K1, K2, ..., K m ), m represents the total number of groups, and the maximum total fit Kmax=max(K jh ) The pending words in the corresponding group are used as target keywords.

2. The cloud archive management system according to claim 1, characterized in that: Obtaining the pending vector and comparison vector includes: For the incoming user, convert a single keyword in the corresponding access keyword sequence into a word vector, perform TF-IDF weighted averaging on all the word vectors to obtain a target vector, which is recorded as the undetermined vector; Obtain the target vector of the target user and record it as a comparison vector.

3. The cloud archive management system according to claim 1, characterized in that: When the data partition corresponding to the merged user corresponds to the remaining target users, perform the following steps: The remaining target users corresponding to the data partition corresponding to the merged user are marked as pending users. When the cosine similarity between the comparison vector of any pending user and the pending vector is less than the cosine similarity threshold, the merged user is no longer considered as a merged user.

4. The cloud archive management system according to claim 1, characterized in that: When two or more cosine similarities are greater than the cosine similarity threshold, a prompt message is sent for reporting.

5. The cloud archive management system according to claim 1, characterized in that: When the target user does not exist, the steps in the second partitioning module are executed to open a data partition for the incoming user.

6. The cloud archive management system according to claim 1, characterized in that: When the registration time of the entering user is less than the preset duration, the subsequent steps are not executed, and a prompt message is sent to the preset administrator, who manually opens the data partition of the entering user.

Images