WO2022088674A1 - 跨数据信息知识模态的面向本质计算的差分内容推荐方法 - Google Patents

跨数据信息知识模态的面向本质计算的差分内容推荐方法 Download PDF

Info

Publication number
WO2022088674A1
WO2022088674A1 PCT/CN2021/097477 CN2021097477W WO2022088674A1 WO 2022088674 A1 WO2022088674 A1 WO 2022088674A1 CN 2021097477 W CN2021097477 W CN 2021097477W WO 2022088674 A1 WO2022088674 A1 WO 2022088674A1
Authority
WO
WIPO (PCT)
Prior art keywords
query
information
data
resources
row
Prior art date
Application number
PCT/CN2021/097477
Other languages
English (en)
French (fr)
Inventor
段玉聪
樊珂
Original Assignee
海南大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 海南大学 filed Critical 海南大学
Publication of WO2022088674A1 publication Critical patent/WO2022088674A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/22Indexing; Data structures therefor; Storage structures
    • G06F16/2282Tablespace storage structures; Management thereof
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24Querying
    • G06F16/245Query processing
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24Querying
    • G06F16/245Query processing
    • G06F16/2458Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2462Approximate or statistical queries
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24Querying
    • G06F16/245Query processing
    • G06F16/2458Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2468Fuzzy queries
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/28Databases characterised by their database models, e.g. relational or object models
    • G06F16/284Relational databases
    • G06F16/285Clustering or classification

Definitions

  • the invention relates to the technical field of content recommendation, and in particular, to a differential content recommendation method oriented to essential computing across data information knowledge modalities.
  • a database is a long-term, organized and shareable collection of data stored in a computer, and is an important part of data resources.
  • a database consists of multiple datasets.
  • Data set is a collection of data, which is an important part of data resources. It usually appears in the form of a table. Each column represents a specific variable and represents different attributes. Each row represents a member corresponding to different specific variables. Value, each value in the table belongs to the data resource. If the content in the table is understood correspondingly, that is, the row and column where the table content is located can be correlated and analyzed, which can express rich semantics and describe the entities and entities in the real world. At this time, the data resources are transformed into information resources.
  • the data of the dataset may include one or more members.
  • the data holder stipulates that third parties cannot directly obtain the data on the data sheet.
  • the third party can only obtain some global statistical data through the query algorithm provided by the database, which maximizes the accuracy of data query and reduces the chance of identifying specific records on the data table. It is difficult for the current recommendation system to be based on such databases. User needs analysis is performed to recommend content, resulting in low recommendation accuracy.
  • the purpose of the present invention is to provide an essential calculation-oriented differential content recommendation method across data information knowledge modalities, so as to overcome or at least partially solve the above-mentioned problems existing in the prior art.
  • An essential computing-oriented differential content recommendation method across data information knowledge modalities comprising the following steps:
  • S3 analyze the data resources and information resources to obtain new information resources, match the push content according to the new information resources of the target user, and push the content to the target user.
  • step S2 specifically includes:
  • step S22 specifically includes:
  • the known data storage location is in the i-th row and the j-th column of the user's personal information data table, execute the queries Q1(i-1,j) and Q1(i,j), take the query results as data resources and information resources, and use Data resources and information resources are fused to obtain new information resources, and the difference between Q1(i-1,j) and Q1(i,j) is calculated and used as the attribute value of the new information resource.
  • the attribute value of the new information resource is the i-th row in the data table.
  • the multi-line records are used as a module to calculate the difference, and the range of the queried data is obtained by performing the difference calculation for many times, and the specific number of rows of the queried data is determined, and the meaning of the attribute value is contacted. Access to new information resources.
  • step S22 specifically includes:
  • the known data storage location is in the i-th row and the j-th column of the user's personal information data table, execute the queries Q1(i-1,j) and Q1(i,j), take the query results as data resources and information resources, and use Data resources and information resources are fused to obtain new information resources, and the difference between Q1(i-1,j) and Q1(i,j) is calculated and used as the attribute value of the new information resource.
  • the attribute value of the new information resource is the i-th row in the data table.
  • the step S22 specifically includes:
  • the selected row record is regarded as a module, and the rows in the module are discontinuous, and Q2 is executed to query the sum of the attribute values of the corresponding row. Perform statistical analysis to obtain data resources and information resources, and analyze and integrate data resources and information resources to obtain new information resources;
  • the queried data is in the j-th column, analyze the restriction rule information , if a single record is allowed to be queried, execute a single-line query Q2(i,j) to obtain the attribute values of all users in the interval, add and sum the attribute values of the records in the interval, obtain relevant knowledge resources, obtain data resources and information resources, and analyze the data Analysis and fusion of resources and information resources to obtain new information resources,
  • the query does not allow direct query of a single record and can only query multiple rows of records with a continuous and fixed number of rows, execute the query Q2 ⁇ (i,j),n ⁇ to count the row interval.
  • step S22 specifically includes:
  • the query is made according to the known attribute value information, which includes: if the specific attribute value of the target user is known or unknown, then each row of column j will be queried.
  • Execute the Q3(i,j) query compare the Q3 query results of the target user's row with the Q3 query results of other rows, and filter all record attribute values according to the comparison results to obtain new information resources;
  • the specific attribute value further determines the row where the target user record is located.
  • step S22 specifically includes:
  • query according to the known attribute value information which specifically includes: if all possible discrete attribute values in the jth column are known, execute Q3(i,j) query to obtain data resources and information resources;
  • the queried data is in the jth column, according to the known attributes
  • step S22 specifically includes:
  • the known data storage location is in the i-th row and the j-th column of the user's personal information data table
  • query based on the known attribute value information which specifically includes: if it is known that the attribute values of multiple records in the j-th column are the i-th row and the j-th row
  • For the attribute values displayed in column j first execute Q4 ⁇ j,“Value 1 ”,“Value 2 ”,...,“Value n ” ⁇ to obtain the total number of records with different attribute values, and then execute Q3(i,j) query Classify the results to obtain data resources and information resources. If all possible attribute values in column j are unknown and the specific attribute values in row i and column j of the target user are unknown, perform fuzzy classification by executing Q3 query row by row to obtain new information. information resource;
  • the queried data is in the jth column, according to the known attributes Querying the value information, which specifically includes: if the attribute value has preset options and the specific attribute value of the target user is unknown, execute the Q3(i,j) query on the jth column of the ath row to the bth row, and perform a fuzzy query according to the query result.
  • the Q4(j, "Value") query is further executed, and the query result is compared with the result obtained by performing the Q3(i,j) query in the jth column of row a to row b. For access to new information resources.
  • step S22 specifically includes:
  • the queried data is in the jth column, according to the known attributes Querying the value information, which specifically includes: if there is a set rule for the attribute value and the specific attribute value of the target user is unknown, execute the Q3(i,j) query on the jth column of the ath row to the bth row, and perform a fuzzy query according to the query result. Classification, if the specific attribute value of the target user is known, the Q4(j, "Value”) query is further executed, and the obtained result is compared with the Q3(i,j) query result to obtain new information resources.
  • the invention provides an essential calculation-oriented differential content recommendation method across data information knowledge modalities.
  • the database restricts query functions and queryable content, it can be based on different query functions and statistical information content disclosed by the database.
  • the target user's data resources and information resources and based on the analysis of data resources and information resources to obtain new information resources, and push the corresponding push data to the target users according to the new information resources, so as to realize the difference in the case of incomplete public data in the privacy database Content push, improve the accuracy of push content.
  • FIG. 1 is a schematic overall flow diagram of a method for recommending differential content based on essential computing across data information knowledge modalities according to an embodiment of the present invention.
  • the present invention provides an essential calculation-oriented differential content recommendation method across data information knowledge modalities, and the method includes the following steps:
  • Data resources are discrete elements obtained by direct observation, which have no meaning without context and are not associated with a specific purpose of human beings. Data resources express the attribute content of a single entity. Information resources record human behavior and are used to mine, analyze, and express the interaction between two entities. An entity can be either another person or an objectively existing thing. Information resources are related to a specific purpose of human beings. Infer the relationship between two entities through purpose. The following will also involve knowledge resources. Knowledge resources are derived from data resources and information resources through structured and formalized derivation and deduction. Knowledge resources abstract entity relationships on the basis of information resources.
  • S3 analyze the data resources and information resources to obtain new information resources, match the push content according to the new information resources of the target user, and push the content to the target user.
  • step S2 specifically includes:
  • the query function includes Q1(i,j), Q2 ⁇ (i1, j1 ),( i2 , j2 ),... ⁇ , Q3( i,j) and Q4 ⁇ j,"Value 1 ",”Value 2 ",...,”Value n " ⁇ , Q1(i,j) is used to query the first row i of the jth column in the user's personal information data table
  • the partial sum of ) is used to query the total number of records with the same attribute value as the i-th row, j-th column in the user personal information data table
  • Q4 is used to query the total number of records in the user personal information data table j column with the same attribute value, If there are more than one, the total number of records corresponding to each specified attribute value is sequentially output; the restriction rule information is used to describe the query restriction of the corresponding push mode.
  • a hospital has a medical record database, and each record in a certain data table in the database is represented as (Name, X, Y, Z, ...), where X , Y, Z are specific variables, X represents whether the patient has gastritis, and the attribute value is represented by a Boolean value, 1 represents gastritis, 0 represents no gastritis; Y represents the height of the patient, and the size is represented by a numerical value; Z represents the attending doctor of the patient, It is represented by a string; P represents a prescription, and a number represents the drug code. Part of the data sheet is shown in Table 1.
  • the hospital database does not directly publish the specific content of the data table. It is stipulated that the pusher can only query the statistical values in the data table through a specific form of query function. Differential content push of the data table.
  • step S22 specifically includes:
  • the known data storage location information is the data table data storage information already known by the pusher. If the known data storage location is in the i-th row and the j-th column of the user's personal information data table, execute the queries Q1(i-1,j) and Q1(i,j), take the query results as data resources and information resources, and use Data resources and information resources are fused to obtain new information resources, and the difference between Q1(i-1,j) and Q1(i,j) is calculated and used as the attribute value of the new information resource.
  • the attribute value of the new information resource is the i-th row in the data table.
  • specific personal information cannot be directly obtained by performing the above difference calculation, the scope of personal information can be narrowed down.
  • the pusher believes that the probability of the patient Dell suffering from gastritis is 50% without the relevant operation. Matching and pushing of recommended content.
  • the multi-line records will be used as a module to calculate the difference, and the difference will be calculated by multiple times.
  • Obtain the range of the queried data determine the specific row number of the queried data, and obtain new information resources by contacting the meaning of the attribute value. For example, there are a large number of records in the data table, the value of the attribute that needs to be screened is a Boolean value, and the number of 0 and 1 values in the records is very different. For example, only 10 records in 500 records have the attribute value of 1.
  • step S22 specifically includes:
  • the known data storage location is in the i-th row and the j-th column of the user's personal information data table, execute the queries Q1(i-1,j) and Q1(i,j), take the query results as data resources and information resources, and use Data resources and information resources are fused to obtain new information resources, and the difference between Q1(i-1,j) and Q1(i,j) is calculated and used as the attribute value of the new information resource.
  • the attribute value of the new information resource is the i-th row in the data table.
  • the concrete attribute value of the jth column For example, the record of the patient Dell is in the seventh row.
  • a single-line difference calculation is performed on the user's personal information data table line by line, and new information resources are obtained by associating with a specific user. For example, in the attribute column Y in the data table, start the summation calculation from the first row, and then subtract it from the previous summation result to obtain the specific height data of different patients.
  • the step S22 specifically includes:
  • the known data storage location is in the i-th row and the j-th column of the user's personal information data table, analyze the restriction rule information. If a single record is allowed to be queried, execute a single-line query Q2(i,j) to obtain the new information resources of the target user. When a single record is allowed to be queried, the partial sum of the specified row and column in the Q2(i,j) query data table is the attribute value corresponding to the single record.
  • the pusher only needs to execute the single-line query Q2(i,j) to obtain the attribute values of all users in the interval in the data table, and the attribute value of the target user is one of them, and then the records in the interval can be added to obtain the attribute values. And, combining statistics and other related knowledge resources to obtain more data resources and information resources, not only can analyze the relevant data resources of target users to obtain new information resources, but also obtain relevant characteristics of certain groups including target users analyze.
  • the queries Q2 ⁇ (i,j),n ⁇ and Q2 ⁇ (i+1,j),n ⁇ for twice The query performs difference calculation, uses the results as data resources and information resources, and analyzes and fuses data resources and information resources to obtain new information resources.
  • the Q2 ⁇ (i,j),n ⁇ query data table specifies that the column is j, and the number of fixed rows of the starting row i is n. The sum of the attribute values of .
  • the selected row record is regarded as a module, and the rows in the module are discontinuous, and Q2 is executed to query the sum of the attribute values of the corresponding row.
  • Statistical analysis is performed to obtain data resources and information resources, and data resources and information resources are analyzed and fused to obtain new information resources. It is not allowed to directly query a single record and can only query the sum of the record attribute values of a fixed number of rows.
  • the records of different rows selected in each module are summed.
  • the Q2 query obtains the sum of attribute values including the target user's row and others.
  • the queried data is in the j-th column, analyze the restriction rule information , if a single record is allowed to be queried, execute a single-line query Q2(i,j) to obtain the attribute values of all users in the interval, add and sum the attribute values of the records in the interval, obtain relevant knowledge resources, obtain data resources and information resources, and analyze the data Analysis and fusion of resources and information resources to obtain new information resources.
  • the pusher executes the Q2 ⁇ (i,j),n ⁇ to count the row interval.
  • the pusher executes the Q2 ⁇ (i,j),n ⁇ query to perform statistical analysis on the row interval, which can be divided into two situations: one is that the Q2 query contains multiple consecutive row records containing a specific row interval, that is, the Q2 ⁇ ( The row interval of a,j),n ⁇ query is [a,a+n-1], there are a total of n rows, and a+n-1 ⁇ b, that is, the number of rows queried by Q2 is greater than or equal to the interval where the target user is located number of rows, then the statistical summation value obtained by a Q2 query is greater than or equal to the statistical summation value of a specific row interval.
  • the Q2 query continuous multi-line records contain some specific row intervals, that is, the row interval for Q2 ⁇ (a,j),n ⁇ query is [a,a+n-1], and a+n -1 ⁇ b, even if the number of rows queried by Q2 is less than the number of rows in the interval where the target user is located, to obtain all records in the row interval [a,b], it is necessary to perform multiple Q2 queries until the rows can be obtained through the Q2 query All records in the interval [a,b].
  • Q2 ⁇ (2,2),5 ⁇ 3.
  • the pusher If the known data storage location is empty, that is, when the pusher does not know the location of the corresponding record of the target user in the data table, if the Q2 query allows querying a single record, the pusher only needs to continuously execute the single-line query Q2(i,j), and then You can get the attribute value of all rows in the data table.
  • the pusher can query module by module from the first record to the last row to obtain the sum of the attribute values of different modules.
  • the sum of the attribute values of can represent some properties and characteristics of multiple records of this module.
  • the sum of attribute values of all modules is obtained by adding up the sum of attribute values of all records, which can represent some properties and characteristics of a large-scale collective and provide more related resources.
  • the Q2 query does not allow direct query of a single record, it can only query the sum of the j column attribute values of discontinuous multi-line records through a certain algorithm and the total number of rows is fixed.
  • the sequence selected in the second module was 11, 13, 15, 17, until the entire data table was screened.
  • the Q2 query does not allow direct query of a single record, but only allows n different rows to be randomly selected for summation, then the summation result will be associated with data resources and information resources, and relevant analysis will be carried out with the support of statistical knowledge resources, etc. Some qualitative characteristics of the whole group.
  • the step S22 specifically includes:
  • the query is made according to the known attribute value information, which includes: if the specific attribute value of the target user is known or unknown, then each row of column j will be queried.
  • Execute the Q3(i,j) query compare the Q3 query results of the target user's row with the Q3 query results of other rows, and filter the attribute values of all records according to the comparison results to obtain new information resources.
  • the Q3(i,j) query is performed on each row of the boolean attribute j column, the result obtained by performing the Q3 query on the row where the target user is located is obtained by performing the Q3 query on other rows. If the results are the same, it means that the row has the same attribute value as the target user, so the attribute values of all records can be separated, and the attribute values corresponding to all records can be obtained by adding other information resources. This can obtain new information resources about different users, so as to match and push recommended content.
  • the step S22 specifically includes:
  • the known data storage location is in the i-th row and the j-th column of the user's personal information data table, determine whether the attribute value of the queried data is discrete. Discrete means that the attribute value has been set in advance, and each row of records is only recorded when the attribute value is entered. can be selected from these set data values;
  • the Q3(i,j) query is performed row by row from the first row of records to the last row, if the number of different results is equal to the number of different set attribute values, it can be known that all attribute values correspond to The number of records, and the records of the data table can be classified according to the Q3 query results.
  • the records with the same result belong to one category, but the attribute value corresponding to this category of records cannot be determined.
  • the corresponding attribute values of all records cannot be obtained at this time, But it also provides help for matching attribute values; if the number of different results is less than the number of set attribute values, the possible situations are: a.
  • the attribute value is obtained through Q3 query
  • the value of is 0; some attribute values have the same number of occurrences and are mistakenly classified into one category; c.
  • the pusher also knows the specific attribute value of the target user, that is, the attribute value displayed in the i-th row and the j-th column, the pusher can determine the attribute value of this type by querying Q3(i,j), narrowing the confirmation range of other attribute values. .
  • the pusher uses a specific Q3(i,j) query at this time, and the obtained result indicates how many records in the jth column of the entire data table have the attribute value of the th
  • the attribute value displayed in row i, column j if the Q3 query is performed row by row from the first row, the records are classified according to the query results, and the number of attribute values is set to be greater than or equal to the number of types of query results to provide support for identifying attribute values. .
  • the step S22 specifically includes:
  • the known data storage location is in the i-th row and the j-th column of the user's personal information data table
  • query based on the known attribute value information which specifically includes: if it is known that the attribute values of multiple records in the j-th column are the i-th row and the j-th row
  • For the attribute values displayed in column j first execute Q4 ⁇ j,“Value 1 ”,“Value 2 ”,...,“Value n ” ⁇ to obtain the total number of records with different attribute values, and then execute Q3(i,j) query Classify the results to obtain data resources and information resources. If all possible attribute values in column j are unknown and the specific attribute values in row i and column j of the target user are unknown, perform fuzzy classification by executing Q3 query row by row to obtain new information. information resource.
  • the queried data is in the jth column, according to the known attributes Querying the value information, which specifically includes: if the attribute value has preset options and the specific attribute value of the target user is unknown, execute the Q3(i,j) query on the jth column of the ath row to the bth row, and perform a fuzzy query according to the query result.
  • the Q4(j, "Value") query is further executed, and the query result is compared with the result obtained by performing the Q3(i,j) query in the jth column of row a to row b. Comparing the results obtained by acquiring new information resources with the results obtained by performing the Q3(i,j) query in the jth column of row a to row b can narrow the scope of identifying the row sequence of the target user record.
  • the known attribute value information which specifically includes: if the attribute value of the string attribute is set in advance, when the pusher does not know the specific attribute value of the target user, the data Perform Q3(i,j) query on all the records in the table to obtain the total number of records corresponding to the attribute value and fuzzy classification; further, if the pusher knows the specific attribute value of the target user, then perform Q4(j, "Value") Query, and compare the results obtained by Q3(i,j) query with the previously calculated results, which can narrow the range of attribute value matching and identifying the row sequence of the target user record, and at the same time narrow the range of matching other attribute values. .
  • the pusher uses Q4(j,”Value") query, and compares the result with the result obtained by using Q3(i,j) query, the row sequence where the same value is located may be this The record line sequence of the target user, thereby obtaining fuzzy identification and narrowing the effective range of identification.
  • the step S22 specifically includes:
  • the queried data is in the jth column, according to the known attributes Querying the value information, which specifically includes: if there is a set rule for the attribute value and the specific attribute value of the target user is unknown, execute the Q3(i,j) query on the jth column of the ath row to the bth row, and perform a fuzzy query according to the query result. Classification, if the specific attribute value of the target user is known, the Q4(j, "Value”) query is further executed, and the obtained result is compared with the Q3(i,j) query result to obtain new information resources.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computational Linguistics (AREA)
  • Automation & Control Theory (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种跨数据信息知识模态的面向本质计算的差分内容推荐方法,该方法包括以下步骤:S1、获取目标用户基本信息,连接存储有目标用户隐私信息的数据库;S2、根据数据库公开统计内容构造查询函数查询目标用户个人信息数据表,获取数据资源和信息资源;S3、对数据资源和信息资源进行分析获得新信息资源,根据目标用户的新信息资源匹配推送内容并向目标用户进行推送。该方法能够在隐私数据库公开数据不完全的情况下实现差分内容推送,提高推送内容的准确率。

Description

跨数据信息知识模态的面向本质计算的差分内容推荐方法 技术领域
本发明涉及内容推荐技术领域,尤其涉及一种跨数据信息知识模态的面向本质计算的差分内容推荐方法。
背景技术
数据库是长期存储在计算机内、有组织的、可共享的数据集合,是数据资源的重要组成部分,对其中存在的数据可由统一软件进行增、删、改、查等管理和控制行为。数据库由多个数据集构成。数据集是一种由数据所组成的集合,是数据资源的重要组成部分,通常以表格形式出现,每一列代表一个特定变量,表示不同属性,每一行都代表某一成员对应于不同特定变量的值,表格中的每一个值都属于数据资源,若对表格中的内容进行对应理解,即对表格内容所在的行与列进行关联分析,能够表达丰富的语义,描述出现实世界的实体以及实体的不同属性值,此时数据资源转化成为信息资源。对应于行数,数据集的数据可能包括一个或多个成员。当一个受信任机构作为数据持有方持有多个用户的敏感个人信息数据集,例如医疗记录、购买记录、通话记录等,为了保护隐私,数据持有方规定第三方无法直接获取数据表上的数据资源,第三方只能通过数据库提供的查询算法获取一些全局性统计数据,使得数据查询的准确性最大化,同时减少识别数据表上具体记录的机会,目前的推荐系统难以基于此类数据库对用户进行需求分析从而进行内容推荐,导致推荐准确率较低。
发明内容
本发明的目的在于提供一种跨数据信息知识模态的面向本质计算的差分内容推荐方法,以克服或至少部分解决现有技术所存在的上述问题。
一种跨数据信息知识模态的面向本质计算的差分内容推荐方法,包括以下步骤:
S1、获取目标用户基本信息,连接存储有目标用户隐私信息的数据库;
S2、根据数据库公开统计内容构造查询函数查询目标用户个人信息数据表,获取数据资源和信息资源;
S3、对数据资源和信息资源进行分析获得新信息资源,根据目标用户的新信息资源匹配推送内容并向目标用户进行推送。
进一步的,所述步骤S2具体包括:
S21、获取数据库所提供查询函数和限制规则信息,所述查询函数包括Q1(i,j)、Q2{(i 1,j 1),(i 2,j 2),...}、Q3(i,j)和Q4{j,“Value 1”,“Value 2”,...,“Value n”},Q1(i,j)用于查询用户个人信息数据表中第j列前i行的部分总和;Q2{(i 1,j 1),(i 2,j 2),...}用于查询用户个人信息数据表中第i行第j列的部分总和;Q3(i,j)用于查询用户个人信息数据表中与第i行第j列的属性值相同的记录总数;Q4用于查询用户个人信息数据表j列中与指定属性值相同的记录总数,若指定属性值有多个则顺序输出与各个指定属性值对应的记录总数;所述限制规则信息用于描述相应推送方式的查询限制;
S22、根据查询函数、限制规则信息和已知数据存储位置信息,设置查询函数参数并发送至数据库进行查询;
S23、根据查询结果获取数据资源和信息资源。
进一步的,当数据库所提供查询函数为Q1(i,j)时,所述步骤S22具体包括:
判断所查询数据是否为布尔型,若为布尔型则判断已知数据存储位置信息;
若已知数据存储位置在用户个人信息数据表中第i行第j列,则执行查询Q1(i-1,j)和Q1(i,j),将查询结果作为数据资源和信息资源,将数据资源和信息资源融合获得新信息资源,计算Q1(i-1,j)和Q1(i,j)的差值并作为新信息资源属性值,新信息资源属性值为数据表中第i行第j列的具体属性值;
若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,则执行查询Q1(a-1,j)和Q1(b,j),将查询结果作为数据资源和信息资源,将数据资源和信息资源融合获得新信息资源,计算Q1(a-1,j)和Q1(b,j)的差值并作为新信息资源属性值;
若已知数据存储位置信息为空,则将多行记录作为一个模块进行差值计算,通过多次进行差值计算获取所查询数据所在范围,确定所查询数据所在具体行数,联系属性值含义获取新信息资源。
进一步的,当数据库所提供查询函数为Q1(i,j)时,所述步骤S22具体包括:
判断所查询数据是否为数值型,若为数值型则判断已知数据存储位置信息;
若已知数据存储位置在用户个人信息数据表中第i行第j列,则执行查询Q1(i-1,j)和Q1(i,j),将查询结果作为数据资源和信息资源,将数据资源和信息资源融合获得新信息资源,计算Q1(i-1,j)和Q1(i,j)的差值并作为新信息资源属性值,新信息资源属性值为数据表中第i行第j列的具体属性值;
若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,则执行查询Q1(a-1,j)和Q1(b,j),将查询结果作为数据资源和信息资源,将数据资源和信息资源融合获得新信息资源,计算Q1(a-1,j)和Q1(b,j)的差值并作为新信息资源属性值;
若已知数据存储位置信息为空,则对用户个人信息数据表逐行进行单行差值计算,通过关联特定用户获取新信息资源。
进一步的,当数据库所提供查询函数为Q2{(i 1,j 1),(i 2,j 2),...}时,所述步骤S22具体包括:
判断所查询数据是否为布尔型或数值型,若为布尔型或数值型则判断已知数据存储位置信息;
若已知数据存储位置在用户个人信息数据表中第i行第j列,分析限制规则信息,若允许查询单条记录,则执行单行查询Q2(i,j),获取目标用户 的新信息资源,
若不允许查询单条记录且仅能查询连续、固定行数的多行记录,则执行查询Q2{(i,j),n}和Q2{(i+1,j),n},对两次查询进行差值计算,将结果作为数据资源和信息资源,对数据资源和信息资源进行分析融合获得新信息资源,
若不允许直接查询单条记录且仅能查询固定行数的记录属性值之和,则将选取的行记录作为一个模块,模块内行不连续,执行Q2查询相应行的属性值之和,对属性值之和进行统计分析获得数据资源、信息资源,对数据资源和信息资源进行分析融合获得新信息资源;
若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,分析限制规则信息,若允许查询单条记录,则执行单行查询Q2(i,j)获取区间内所有用户的属性值,将区间内记录属性值相加求和,获取相关知识资源获得数据资源和信息资源,对数据资源和信息资源进行分析融合获得新信息资源,
若查询不允许直接查询单条记录且仅能查询连续、固定行数的多行记录,则执行查询Q2{(i,j),n}对行区间进行统计。
进一步的,当数据库所提供查询函数为Q3(i,j)和Q4{j,“Value 1”,“Value 2”,...,“Value n”}时,所述步骤S22具体包括:
判断所查询数据是否为布尔型,若为布尔型则判断已知数据存储位置信息;
若已知数据存储位置在用户个人信息数据表中第i行第j列,根据已知属性值信息进行查询,具体包括:若已知或未知目标用户具体属性值,则对j列每一行都执行Q3(i,j)查询,将目标用户所在行的Q3查询结果与其他行进行Q3查询结果进行对比,根据对比结果对所有记录属性值进行筛选,获得新信息资源;
若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,根据属性值信息进行查询,具体包括:若已知第j列中所有可能出现的属性值,则从Q3(a,j)开始逐行进 行查询,直到i=b,将查询结果作为新信息资源,若已知具体属性值则进一步确定目标用户记录所在行。
进一步的,当数据库所提供查询函数为Q3(i,j)和Q4{j,“Value 1”,“Value 2”,...,“Value n”}时,所述步骤S22具体包括:
判断所查询数据是否为数值型,若为数值型则判断已知数据存储位置信息;
若已知数据存储位置在用户个人信息数据表中第i行第j列,判断所查询数据属性值是否离散;
若离散,根据已知属性值信息进行查询,具体包括:若已知第j列所有可能出现的离散属性值,执行Q3(i,j)查询获取数据资源和信息资源;
若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,根据已知属性值信息进行查询,具体包括:若已知第j列所有可能出现的离散属性值,则从Q3(a,j)开始逐行进行查询,直到i=b,将查询结果作为新信息资源,若已知具体属性值则执行Q4(j,“Value”)查询,将查询结果与多行记录进行Q3查询的结果相匹配获取数据资源和信息资源;
若已知数据存储位置为空,则先通过执行Q4{j,“Value 1”,“Value 2”,...,“Value n”}查询不同属性值记录总数,再执行Q3(i,j)查询对结果进行分类以获取数据资源和信息资源。
进一步的,当数据库所提供查询函数为Q3(i,j)和Q4{j,“Value 1”,“Value 2”,...,“Value n”}时,所述步骤S22具体包括:
判断所查询数据是否为字符型,若为字符型则判断已知数据存储位置信息;
若已知数据存储位置在用户个人信息数据表中第i行第j列,根据已知属性值信息进行查询,具体包括:若已知第j列共有多个记录的属性值是第i行第j列显示的属性值,则先执行Q4{j,“Value 1”,“Value 2”,...,“Value n”}获得不同属性值的记录总数,再执行Q3(i,j)查询对结果进行分类以 获取数据资源和信息资源,若未知第j列所有可能出现的属性值且未知目标用户第i行第j列的具体属性值,则通过逐行执行Q3查询进行模糊分类获取新信息资源;
若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,根据已知属性值信息进行查询,具体包括:若属性值具有预设选项且未知目标用户具体属性值时,对第a行至第b行的第j列执行Q3(i,j)查询,根据查询结果进行模糊分类,若已知目标用户具体属性值则进一步执行Q4(j,“Value”)查询,并将查询结果与a行至b行的第j列进行Q3(i,j)查询得到的结果进行比对获取新信息资源。
进一步的,当数据库所提供查询函数为Q3(i,j)和Q4{j,“Value 1”,“Value 2”,...,“Value n”}时,所述步骤S22具体包括:
判断所查询数据是否为代码型,若为代码型则判断已知数据存储位置信息;
若已知数据存储位置在用户个人信息数据表中第i行第j列,根据已知属性值信息进行查询,具体包括:若属性值存在设定规则且未知具体属性值或未知命名规则,执行Q3(i,j)查询进行模糊分类,若已知目标用户特定属性值则执行Q4{j,“Value 1”,“Value 2”,...,“Value n”}查询分析获取具体属性值;
若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,根据已知属性值信息进行查询,具体包括:若属性值存在设定规则且未知目标用户具体属性值时,对第a行至第b行的第j列执行Q3(i,j)查询,根据查询结果进行模糊分类,若已知目标用户具体属性值,则进一步执行Q4(j,“Value”)查询,将获得结果与执行Q3(i,j)查询结果进行比对,获取新信息资源。
进一步的,在根据已知数据存储位置信息和已知属性值信息只能进行模糊分类时,引入与相应数据资源相关的数据资源、信息资源或知识资源进行同模态或跨模态关联融合。
与现有技术相比,本发明的有益效果是:
本发明所提供的一种跨数据信息知识模态的面向本质计算的差分内容推荐方法,在数据库限制查询函数和可查询内容的情况下,能够基于不同的查询函数和数据库公开的统计信息内容获取目标用户的数据资源和信息资源,并基于数据资源和信息资源分析获取新信息资源,根据新信息资源匹配相应的推送数据向目标用户进行推送,从而在隐私数据库公开数据不完全的情况下实现差分内容推送,提高推送内容的准确率。
附图说明
为了更清楚地说明本发明实施例中的技术方案,下面将对实施例描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本发明的优选实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。
图1是本发明实施例提供的一种跨数据信息知识模态的面向本质计算的差分内容推荐方法整体流程示意图。
具体实施方式
以下结合附图对本发明的原理和特征进行描述,所列举实施例只用于解释本发明,并非用于限定本发明的范围。
参照图1,本发明提供一种跨数据信息知识模态的面向本质计算的差分内容推荐方法,所述方法包括以下步骤:
S1、获取目标用户基本信息,连接存储有目标用户隐私信息的数据库。
S2、根据数据库公开统计内容构造查询函数查询目标用户个人信息数据表,获取数据资源和信息资源。数据资源是由直接观察得到的离散元素,在没有上下文的情况下不具有任何意义,不与人类的某个特定目的相关联,数据资源表达单个实体的属性内容。信息资源记录人类的行为,用于挖掘、分 析、表达两个实体之间的交互关系,实体既可以是另一个人,也可以是客观存在的事物,信息资源与人类的某个特定目的相关,透过目的去推断两个实体之间的关系。下文还将涉及到知识资源,知识资源由数据资源和信息资源经过结构化形式化的推导演绎得到,知识资源在信息资源的基础上对实体关系进行了抽象化的归纳总结。
S3、对数据资源和信息资源进行分析获得新信息资源,根据目标用户的新信息资源匹配推送内容并向目标用户进行推送。
其中所述步骤S2具体包括:
S21、获取数据库所提供查询函数和限制规则信息,所述查询函数包括Q1(i,j)、Q2{(i 1,j 1),(i 2,j 2),...}、Q3(i,j)和Q4{j,“Value 1”,“Value 2”,...,“Value n”},Q1(i,j)用于查询用户个人信息数据表中第j列前i行的部分总和;Q2{(i 1,j 1),(i 2,j 2),...}用于查询用户个人信息数据表中第i行第j列的部分总和;Q3(i,j)用于查询用户个人信息数据表中与第i行第j列的属性值相同的记录总数;Q4用于查询用户个人信息数据表j列中与指定属性值相同的记录总数,若指定属性值有多个则顺序输出与各个指定属性值对应的记录总数;所述限制规则信息用于描述相应推送方式的查询限制。
S22、根据查询函数、限制规则信息和已知数据存储位置信息,设置查询函数参数并发送至数据库进行查询。
S23、根据查询结果获取数据资源和信息资源。
以下结合一示例以说明本发明技术方案:假设某医院拥有一个医疗记录数据库,数据库中的某一数据表中的每条记录表示为(Name,X,Y,Z,...),其中X、Y、Z为特定变量,X表示病人是否患有胃炎,属性值用布尔值表示,1表示有胃炎,0表示没有胃炎;Y表示病人身高,用数值表示大小;Z表示病人的主治医生,用字符串表示;P表示处方,用数字表示药品代码。部分数据表如表1所示。
表1医疗记录数据库部分数据表
Name X(Yes/No) Y Z P
Emory 0 180 Dr.Chen 1002
German 1 173 Dr.Chen 2003
Marci 0 159 Dr.Li 4001
Damion 0 186 Dr.Liu 4006
Ronald 1 177 Dr.Chen 2003
Karrie 1 167 Dr.Li 4001
Dell 1 181 Dr.Shen 2003
Vince 0 170 Dr.Chen 1002
Aldo 0 155 Dr.Shen 1004
Ryan 0 180 Dr.Li 3008
该医院数据库不直接公布数据表的具体内容,规定推送方只能通过特定形式的查询函数查询数据表中的统计值,接下来基于该例子对步骤S21中的不同查询函数进行说明如何实现基于该数据表的差分内容推送。
当数据库所提供查询函数为Q1(i,j)时,所述步骤S22具体包括:
判断所查询数据是否为布尔型,若为布尔型则判断已知数据存储位置信息。所述已知数据存储位置信息为推送方已经知晓的数据表数据存储信息。若已知数据存储位置在用户个人信息数据表中第i行第j列,则执行查询Q1(i-1,j)和Q1(i,j),将查询结果作为数据资源和信息资源,将数据资源和信息资源融合获得新信息资源,计算Q1(i-1,j)和Q1(i,j)的差值并作为新信息资源属性值,新信息资源属性值为数据表中第i行第j列的具体属性值。例如患者Dell的记录,计算两个查询的差别,即计算Q1(7,2)-Q1(6,2)=4-3=1,融合分析得到的新信息资源是患者Dell的肺炎患病情况,差值是1表示Del l是肺炎患者,若差值为0则表示患者不是肺炎患者,根据Dell的患病情况匹配肺炎治疗药物、用具或者预防用品,从而实现差分内容推荐。
若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,则执行查询Q1(a-1,j)和Q1(b,j),将查询结果作为数据资源和信息资源,将数据资源和信息资源融合获得新信息资源,计算Q1(a-1,j)和Q1(b,j)的差值并作为新信息资源属性值。例如推送方已知患者Dell的记录所属行数i∈[5,8],则推送方先执行两个查询Q1(4,2)和Q1(8,2),分别计算第二列前4行和前8行的总和,然后计算两个查询的差值,得到Q1(8,2)-Q1(4,2)=4-1=3,将已有的数据资源和信息资源融合,分析得到新信息资源为“第5行到第8行共4位患者中共有3人患有肺炎”,进行以上差值计算虽然无法直接得到特定个人信息,但可将个人信息的范围缩小,在此例中,推送方在未经相关操作时,认为患者Dell患胃炎的可能性是50%,进行多行差异的差值计算后,认为患者Dell患胃炎的可能性则会上升,据此同样可以进行推荐内容的匹配和推送。
若已知数据存储位置信息为空,即推送方并不知道目标用户的相关医疗记录在数据表中的位置时,则将多行记录作为一个模块进行差值计算,通过多次进行差值计算获取所查询数据所在范围,确定所查询数据所在具体行数,联系属性值含义获取新信息资源。例如数据表中存在大量记录,需要进行筛查的属性其取值是布尔值,且记录中0和1取值的数目差距很大,比如500条记录中只有10个记录的属性值为1,此时可从第一条记录、最后一条记录甚至是随意某一行记录开始,以50条记录为一个模块进行差值计算,若从第一条记录开始,则执行查询Q1(50,2)得到前50行共有几个属性值为1的记录,接下来执行查询Q1(100,2)、Q1(100,2)、Q1(50,2)可得51行至100行属性值为1的记录,继续进行多行差异的差值计算,可得到属性值为1的记录其大致所在范围,后期通过修改模块大小,可以确定属性值为1的记录的具体行数,最后联系属性值的具体含义,可以得到特定用户信息,从而根据 特定用户信息进行推荐内容的匹配和推送。
当数据库所提供查询函数为Q1(i,j)时,所述步骤S22具体包括:
判断所查询数据是否为数值型,若为数值型则判断已知数据存储位置信息;
若已知数据存储位置在用户个人信息数据表中第i行第j列,则执行查询Q1(i-1,j)和Q1(i,j),将查询结果作为数据资源和信息资源,将数据资源和信息资源融合获得新信息资源,计算Q1(i-1,j)和Q1(i,j)的差值并作为新信息资源属性值,新信息资源属性值为数据表中第i行第j列的具体属性值。例如患者Dell的记录在第七行,推送方想要知道Dell的具体身高数据,只需要执行两个查询Q1(6,3)和Q1(7,3),即分别计算第三列前6行和前7行的总和,然后计算两个查询的差别,即计算Q1(7,3)-Q1(6,3)=1223-1042=181,融合分析得到的新信息资源是患者Dell的具体身高为181厘米,从而据此进行推荐内容的匹配和推送。
若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,则执行查询Q1(a-1,j)和Q1(b,j),将查询结果作为数据资源和信息资源,将数据资源和信息资源融合获得新信息资源,计算Q1(a-1,j)和Q1(b,j)的差值并作为新信息资源属性值。例如推送方已知患者Dell的记录所属行数i∈[5,8],则推送方先执行两个查询Q1(4,3)和Q1(8,3),分别计算第二列前4行和前8行的总和,然后计算两个查询的差值,得到Q1(8,3)-Q1(4,3)=1393-698=695,将已有的数据资源和信息资源融合,分析得到新信息资源为“第5行到第8行共4位病人的身高数据总和为695厘米”。进行以上差值计算虽然无法直接得到特定个人信息,但可将个人信息的范围缩小,从而提高差分内容推荐的精确率。
若已知数据存储位置信息为空,则对用户个人信息数据表逐行进行单行差值计算,通过关联特定用户获取新信息资源。例如,在数据表中属性列Y中,从第一行开始进行求和计算,再与上一个求和结果相减,即得不同病人 的具体身高数据。
当数据库所提供查询函数为Q2{(i1,j1),(i2,j2),...}时,所述步骤S22具体包括:
判断所查询数据是否为布尔型或数值型,若为布尔型或数值型则判断已知数据存储位置信息;
若已知数据存储位置在用户个人信息数据表中第i行第j列,分析限制规则信息,若允许查询单条记录,则执行单行查询Q2(i,j),获取目标用户的新信息资源。允许查询单条记录时,Q2(i,j)查询数据表中指定行、列的部分总和即为单条记录对应的属性值。推送方只需要执行单行查询Q2(i,j)获得数据表中该区间内所有用户的属性值,目标用户的属性值为其中之一,然后可将该区间内的记录进行属性值相加求和,结合统计等相关知识资源得到更多的数据资源、信息资源,不仅可以对目标用户的相关数据资源进行分析从而获得新信息资源,还可得到包括目标用户在内的某些群体的相关特征分析。
若不允许查询单条记录且仅能查询连续、固定行数的多行记录,则执行查询Q2{(i,j),n}和Q2{(i+1,j),n},对两次查询进行差值计算,将结果作为数据资源和信息资源,对数据资源和信息资源进行分析融合获得新信息资源。不允许查询单条记录且仅能查询连续、固定行数的多行记录时,Q2{(i,j),n}查询数据表中指定列为j、起始行为i的固定数量行数为n的属性值总和。
若不允许直接查询单条记录且仅能查询固定行数的记录属性值之和,则将选取的行记录作为一个模块,模块内行不连续,执行Q2查询相应行的属性值之和,对属性值之和进行统计分析获得数据资源、信息资源,对数据资源和信息资源进行分析融合获得新信息资源。不允许直接查询单条记录且仅能查询固定行数的记录属性值之和,总行数固定时,记为Q2(j,n,F(x,t)),F(x,t)是能计算出所选行序的循环算法,x表示算法选出的第几个行记录,t表示 当前循环次数,结束循环的条件是选取的行记录为空,n表示进行一轮循环得到的行序数量,F(0)表示循环算法从第几行开始,可随意选取,每一轮循环选出的n行作为一个模块进行求和,每个模块的和单独显示。若已知目标用户在数据表中所在行序为4,推送方查询Q2(2,4,F(x,t)),表示利用F(x,t)进行模块内行数为4的循环,对每个模块内选取的不同行记录进行求和,已知选取不同行的算法为F(x,t)=F(0)+2x,x=n(t-1)+i,i=1,2,...,n,F(0)=1表示第一轮循环从第一行之后开始计算行序。则推送方进行计算F(x,t)=F(0)+2x=4,推出当F(0)=0时,x=2=4(1-1)+2,即t=1,i=2,F(2,1)=4,表示从第一行开始选取,进行第一轮循环第二个被选的行即为数据表中第四行,第一个模块中选取行序为2,4,6,8,和为2。除此外根据不同F(0)还有其他推送方式,得到的求和结果也不同。此时Q2查询得到包括目标用户所在行以及其他人的属性值之和,虽然无法直接识别特定数据资源,但可确定一个范围,完成模糊识别,提高推送准确率。
若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,分析限制规则信息,若允许查询单条记录,则执行单行查询Q2(i,j)获取区间内所有用户的属性值,将区间内记录属性值相加求和,获取相关知识资源获得数据资源和信息资源,对数据资源和信息资源进行分析融合获得新信息资源。
若查询不允许直接查询单条记录且仅能查询连续、固定行数的多行记录,则执行查询Q2{(i,j),n}对行区间进行统计。此情况下,推送方执行Q2{(i,j),n}查询对行区间进行统计分析,可分为两种情形:一是Q2查询连续多行记录包含特定行区间,即进行Q2{(a,j),n}查询的行区间为[a,a+n-1],总共有n行,且a+n-1≥b,即使用Q2查询到的行数大于等于目标用户所在区间的行数,则进行一次Q2查询得到的统计求和值大于或等于特定行区间统计求和值,结合已有知识资源能解释说明多行连续记录这一群体的某些特征,缩小识别目标用户数据资源的范围;二是Q2查询连续多行记录包含部分特定行 区间,即进行Q2{(a,j),n}查询的行区间为[a,a+n-1],且a+n-1<b,即使用Q2查询到的行数小于目标用户所在区间的行数,则要获取行区间[a,b]内所有记录,需要进行多次Q2查询,直到通过Q2查询可以获取行区间[a,b]内所有记录。例如推送方已知目标用户的相关医疗记录在行区间[2,5],多行模块查询Q2{(i,j),n}规定n=5,此种情形区间总行数小于n,因此只需要进行一次Q2查询,即Q2{(2,2),5}=3,得到的属性值总和虽然不能恰好表现行区间的总和值,但也为识别行区间提供一些数据资源、信息资源,缩小了识别范围。若推送方已知目标用户的相关医疗记录在行区间[2,7],执行查询Q2{(i,j),n}规定n=4,此种情形区间总行数大于n,因此需要进行两次Q2查询才能包含区间内所有行记录,即Q2{(2,2),4}=2,Q2{(6,2),4}=2,将两次查询结果相加得到包含行区间和其他所有记录的总和值,缩小行区间的识别范围。
若已知数据存储位置为空,即推送方不知道目标用户相应记录在数据表中位置时,若Q2查询允许查询单条记录,则推送方只需要不断执行单行查询Q2(i,j),就可得到数据表中所有行的属性值。
若Q2查询不允许直接查询单条记录,只能查询连续的多行记录且行数固定,则推送方可从第一条记录开始逐模块查询,直到最后一行,得到不同模块的属性值之和,模块的属性值之和可以表示该模块多条记录的某些性质和特征。将所有模块的属性值之和相加得到全部记录的属性值之和,可以表示一个大规模集体的某些性质和特征,提供更多相关资源。
若Q2查询不允许直接查询单条记录,只能通过某种算法查询不连续的多行记录的j列属性值之和且总行数固定,若不知道所要识别的病人所在行序,例如推送方进行Q2(2,4,F(x,t))查询,选取不同行的算法F(x,t)=F(0)+2x,x=n(t-1)+i。若不知目标用户所在行序,则推送方进行查询Q2(2,4,F(x,t)),得到F(1,1)=3,F(2,1)=5,F(3,1)=7,F(4,1)=9,即第一轮循环所选行序为3,5,7,9,将其作为第一模块,其属性值之和为2,说明这四位病人中有两位患有胃炎,同理第二模块所选序列为11,13,15,17,直到筛选完整个数据表。查询结果虽然无法直接识别出特定人员的数据资源,但也缩小了识 别范围,为得到真实数据资源提供了相应支持。
若Q2查询不允许直接查询单条记录,只允许随机选取n个不同行进行求和,则将求和结果再关联融合数据资源、信息资源,在统计知识资源等的支持下进行相关分析,得到关于整个群体的一些性质特征。
作为一个示例,当数据库所提供查询函数为Q3(i,j)和Q4{j,“Value 1”,“Value 2”,...,“Value n”}时,所述步骤S22具体包括:
判断所查询数据是否为布尔型,若为布尔型则判断已知数据存储位置信息;
若已知数据存储位置在用户个人信息数据表中第i行第j列,根据已知属性值信息进行查询,具体包括:若已知或未知目标用户具体属性值,则对j列每一行都执行Q3(i,j)查询,将目标用户所在行的Q3查询结果与其他行进行Q3查询结果进行对比,根据对比结果对所有记录属性值进行筛选,获得新信息资源。无论已知还是不知目标用户的具体属性值,若对布尔值属性j列的每一行都进行Q3(i,j)查询,将目标用户所在行进行Q3查询得到的结果与其他行进行Q3查询得到的结果进行对比,若结果相同则表示该行与目标用户有相同属性值,由此可将所有记录的属性值分开,再加上其他信息资源补充,就可得到全部记录对应的属性值,据此可以获得关于不同用户的新信息资源,从而进行推荐内容的匹配和推送。
若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,根据属性值信息进行查询,具体包括:若已知第j列中所有可能出现的属性值,则从Q3(a,j)开始逐行进行查询,直到i=b,将查询结果作为新信息资源,例如布尔值属性的属性值只有两种,则进行Q3查询后的结果也只有两种或一种;结果有两种时可以按属性值将不同记录分开,结果有一种时表示数据表中不同属性值的记录个数相同,可将其作为新信息资源。若已知具体属性值则进一步确定目标用户记录所在行。
作为一个示例,当数据库所提供查询函数为Q3(i,j)和Q4{j,“Value 1”, “Value 2”,...,“Value n”}时,所述步骤S22具体包括:
判断所查询数据是否为数值型,若为数值型则判断已知数据存储位置信息;
若已知数据存储位置在用户个人信息数据表中第i行第j列,判断所查询数据属性值是否离散,离散表示属性值是提前设定过的,每一行记录在录入属性值时都只能从这些被设定好的数据值中选取;
若离散,根据已知属性值信息进行查询,具体包括:若已知第j列所有可能出现的离散属性值,执行Q3(i,j)查询获取数据资源和信息资源,具体包括:若推送方已知第j列中所有可能出现的离散属性值,此时使用特定的Q3(i,j)查询,得到的结果表示在整个数据表的第j列共有多少个记录的属性值是第i行第j列显示的属性值,若从第一行记录开始逐行进行Q3(i,j)查询直到最后一行,若不同结果的数量等于不同设定属性值的数量,则可知全部属性值对应的记录个数,并且可将数据表的记录按Q3查询结果进行分类,结果相同的记录属于一类,但无法确定这一类记录对应的属性值,此时虽然无法得到所有记录的对应属性值,但也为匹配属性值提供了帮助;若不同结果的数量小于设定属性值的数量,则可能情形有:a.数据表存在没有被选择的设定属性值,即该属性值经过Q3查询得到的值为0;存在某些属性值的出现次数相同,被误归为一类;c.以上两种情况同时存在,此时可大致进行属性值分类,但准确度不高。若推送方还知道目标用户的具体属性值,即第i行第j列显示的属性值,则推送方通过Q3(i,j)查询就可确定该类属性值,缩小其他属性值的确认范围。若推送方不知第j列中所有可能出现的离散属性值,此时使用特定的Q3(i,j)查询,得到的结果表示在整个数据表的第j列共有多少个记录的属性值是第i行第j列显示的属性值,若从第一行逐行进行Q3查询,按查询结果对记录进行分类,设定属性值的数量大于或等于查询结果的种类数量,为识别属性值提供支持。
若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,根据已知属性值信息进行查询, 具体包括:若已知第j列所有可能出现的离散属性值,则从Q3(a,j)开始逐行进行查询,直到i=b,将查询结果作为新信息资源,若已知具体属性值则执行Q4(j,“Value”)查询,将查询结果与多行记录进行Q3查询的结果相匹配获取数据资源和信息资源。
若已知数据存储位置为空,则先通过执行Q4{j,“Value 1”,“Value 2”,...,“Value n”}查询不同属性值记录总数,再执行Q3(i,j)查询对结果进行分类以获取数据资源和信息资源,这要求推送方已知第j列中所有可能出现的离散属性值。上述方法对于数据属性值连续或离散的情形均可用于缩小识别范围。
作为一个示例,当数据库所提供查询函数为Q3(i,j)和Q4{j,“Value 1”,“Value 2”,...,“Value n”}时,所述步骤S22具体包括:
判断所查询数据是否为字符型,若为字符型则判断已知数据存储位置信息;
若已知数据存储位置在用户个人信息数据表中第i行第j列,根据已知属性值信息进行查询,具体包括:若已知第j列共有多个记录的属性值是第i行第j列显示的属性值,则先执行Q4{j,“Value 1”,“Value 2”,...,“Value n”}获得不同属性值的记录总数,再执行Q3(i,j)查询对结果进行分类以获取数据资源和信息资源,若未知第j列所有可能出现的属性值且未知目标用户第i行第j列的具体属性值,则通过逐行执行Q3查询进行模糊分类获取新信息资源。
若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,根据已知属性值信息进行查询,具体包括:若属性值具有预设选项且未知目标用户具体属性值时,对第a行至第b行的第j列执行Q3(i,j)查询,根据查询结果进行模糊分类,若已知目标用户具体属性值则进一步执行Q4(j,“Value”)查询,并将查询结果与a行至b行的第j列进行Q3(i,j)查询得到的结果进行比对获取新信息资源,得到的结果与a行至b行的第j列进行Q3(i,j)查询得到的结果进行比对, 可缩小识别目标用户记录所在行序的范围。
若已知数据存储位置信息为空,根据已知属性值信息进行查询,具体包括:若字符串属性的属性值是提前设定好的,当推送方不知目标用户的具体属性值时,对数据表的全体记录进行Q3(i,j)查询,得到对应属性值的记录总数以及模糊分类;进一步地,若推送方已知目标用户的具体属性值,则再进行Q4(j,“Value”)查询,得到的结果与之前计算得到的结果进行Q3(i,j)查询得到的结果进行比对,可缩小属性值匹配和识别目标用户记录所在行序的范围,同时缩小其他属性值匹配的范围。
若字符串属性的属性值不是提前设定好的,当推送方不知目标用户的具体属性值时,不可使用Q4{j,“Value 1”,“Value 2”,...,“Value n”}进行查询,只能使用Q3(i,j)查询全体记录,得到对应属性值出现次数,无法确定目标用户记录的行序和其他相关信息。当推送方已知目标用户的具体属性值时,采用Q4(j,“Value”)查询,将结果与使用Q3(i,j)查询得到的结果进行对比,相同值所在的行序可能就是该目标用户的记录行序,由此得到模糊识别并缩小识别的有效范围。
作为一个示例,当数据库所提供查询函数为Q3(i,j)和Q4{j,“Value 1”,“Value 2”,...,“Value n”}时,所述步骤S22具体包括:
判断所查询数据是否为代码型,若为代码型则判断已知数据存储位置信息。
若已知数据存储位置在用户个人信息数据表中第i行第j列,根据已知属性值信息进行查询,具体包括:若属性值存在设定规则且未知具体属性值或未知命名规则,执行Q3(i,j)查询进行模糊分类,若已知目标用户特定属性值则执行Q4{j,“Value 1”,“Value 2”,...,“Value n”}查询分析获取具体属性值。
若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,根据已知属性值信息进行查询,具体包括:若属性值存在设定规则且未知目标用户具体属性值时,对第a行 至第b行的第j列执行Q3(i,j)查询,根据查询结果进行模糊分类,若已知目标用户具体属性值,则进一步执行Q4(j,“Value”)查询,将获得结果与执行Q3(i,j)查询结果进行比对,获取新信息资源。
若已知数据存储位置信息为空,同样可以参考上述方法缩小识别范围。
在前述实施例的基础上,在根据已知数据存储位置信息和已知属性值信息只能进行模糊分类时,若只依靠原有的数据资源则无法难以基于有效的各模态资源来实现差分内容推荐,此时可引入与相应数据资源相关的数据资源、信息资源或知识资源进行同模态或跨模态关联融合,从而使差分内容的匹配和推送更加准确,效率更高。
以上所述仅为本发明的较佳实施例,并不用以限制本发明,凡在本发明的精神和原则之内,所作的任何修改、等同替换、改进等,均应包含在本发明的保护范围之内。

Claims (10)

  1. 一种跨数据信息知识模态的面向本质计算的差分内容推荐方法,其特征在于,所述方法包括以下步骤:
    S1、获取目标用户基本信息,连接存储有目标用户隐私信息的数据库;
    S2、根据数据库公开统计内容构造查询函数查询目标用户个人信息数据表,获取数据资源和信息资源;
    S3、对数据资源和信息资源进行分析获得新信息资源,根据目标用户的新信息资源匹配推送内容并向目标用户进行推送。
  2. 根据权利要求1所述的一种跨数据信息知识模态的面向本质计算的差分内容推荐方法,其特征在于,所述步骤S2具体包括:
    S21、获取数据库所提供查询函数和限制规则信息,所述查询函数包括Q1(i,j)、Q2{(i 1,j 1),(i 2,j 2),...}、Q3(i,j)和Q4{j,“Value 1”,“Value 2”,...,“Value n”},Q1(i,j)用于查询用户个人信息数据表中第j列前i行的部分总和;Q2{(i 1,j 1),(i 2,j 2),...}用于查询用户个人信息数据表中第i行第j列的部分总和;Q3(i,j)用于查询用户个人信息数据表中与第i行第j列的属性值相同的记录总数;Q4用于查询用户个人信息数据表j列中与指定属性值相同的记录总数,若指定属性值有多个则顺序输出与各个指定属性值对应的记录总数;所述限制规则信息用于描述相应推送方式的查询限制;
    S22、根据查询函数、限制规则信息和已知数据存储位置信息,设置查询函数参数并发送至数据库进行查询;
    S23、根据查询结果获取数据资源和信息资源。
  3. 根据权利要求2所述的一种跨数据信息知识模态的面向本质计算的差分内容推荐方法,其特征在于,当数据库所提供查询函数为Q1(i,j)时,所述步骤S22具体包括:
    判断所查询数据是否为布尔型,若为布尔型则判断已知数据存储位置信 息;
    若已知数据存储位置在用户个人信息数据表中第i行第j列,则执行查询Q1(i-1,j)和Q1(i,j),将查询结果作为数据资源和信息资源,将数据资源和信息资源融合获得新信息资源,计算Q1(i-1,j)和Q1(i,j)的差值并作为新信息资源属性值,新信息资源属性值为数据表中第i行第j列的具体属性值;
    若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,则执行查询Q1(a-1,j)和Q1(b,j),将查询结果作为数据资源和信息资源,将数据资源和信息资源融合获得新信息资源,计算Q1(a-1,j)和Q1(b,j)的差值并作为新信息资源属性值;
    若已知数据存储位置信息为空,则将多行记录作为一个模块进行差值计算,通过多次进行差值计算获取所查询数据所在范围,确定所查询数据所在具体行数,联系属性值含义获取新信息资源。
  4. 根据权利要求2所述的一种跨数据信息知识模态的面向本质计算的差分内容推荐方法,其特征在于,当数据库所提供查询函数为Q1(i,j)时,所述步骤S22具体包括:
    判断所查询数据是否为数值型,若为数值型则判断已知数据存储位置信息;
    若已知数据存储位置在用户个人信息数据表中第i行第j列,则执行查询Q1(i-1,j)和Q1(i,j),将查询结果作为数据资源和信息资源,将数据资源和信息资源融合获得新信息资源,计算Q1(i-1,j)和Q1(i,j)的差值并作为新信息资源属性值,新信息资源属性值为数据表中第i行第j列的具体属性值;
    若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,则执行查询Q1(a-1,j)和Q1(b,j),将查询结果作为数据资源和信息资源,将数据资源和信息资源融合获得新信息资源,计算Q1(a-1,j)和Q1(b,j)的差值并作为新信息资源属性值;
    若已知数据存储位置信息为空,则对用户个人信息数据表逐行进行单行差值计算,通过关联特定用户获取新信息资源。
  5. 根据权利要求2所述的一种跨数据信息知识模态的面向本质计算的差分内容推荐方法,其特征在于,当数据库所提供查询函数为Q2{(i 1,j 1),(i 2,j 2),...}时,所述步骤S22具体包括:
    判断所查询数据是否为布尔型或数值型,若为布尔型或数值型则判断已知数据存储位置信息;
    若已知数据存储位置在用户个人信息数据表中第i行第j列,分析限制规则信息,若允许查询单条记录,则执行单行查询Q2(i,j),获取目标用户的新信息资源,
    若不允许查询单条记录且仅能查询连续、固定行数的多行记录,则执行查询Q2{(i,j),n}和Q2{(i+1,j),n},对两次查询进行差值计算,将结果作为数据资源和信息资源,对数据资源和信息资源进行分析融合获得新信息资源,
    若不允许直接查询单条记录且仅能查询固定行数的记录属性值之和,则将选取的行记录作为一个模块,模块内行不连续,执行Q2查询相应行的属性值之和,对属性值之和进行统计分析获得数据资源、信息资源,对数据资源和信息资源进行分析融合获得新信息资源;
    若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,分析限制规则信息,若允许查询单条记录,则执行单行查询Q2(i,j)获取区间内所有用户的属性值,将区间内记录属性值相加求和,获取相关知识资源获得数据资源和信息资源,对数据资源和信息资源进行分析融合获得新信息资源,
    若查询不允许直接查询单条记录且仅能查询连续、固定行数的多行记录,则执行查询Q2{(i,j),n}对行区间进行统计。
  6. 根据权利要求2所述的一种跨数据信息知识模态的面向本质计算的 差分内容推荐方法,其特征在于,当数据库所提供查询函数为Q3(i,j)和Q4{j,“Value 1”,“Value 2”,...,“Value n”}时,所述步骤S22具体包括:
    判断所查询数据是否为布尔型,若为布尔型则判断已知数据存储位置信息;
    若已知数据存储位置在用户个人信息数据表中第i行第j列,根据已知属性值信息进行查询,具体包括:若已知或未知目标用户具体属性值,则对j列每一行都执行Q3(i,j)查询,将目标用户所在行的Q3查询结果与其他行进行Q3查询结果进行对比,根据对比结果对所有记录属性值进行筛选,获得新信息资源;
    若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,根据属性值信息进行查询,具体包括:若已知第j列中所有可能出现的属性值,则从Q3(a,j)开始逐行进行查询,直到i=b,将查询结果作为新信息资源,若已知具体属性值则进一步确定目标用户记录所在行。
  7. 根据权利要求2所述的一种跨数据信息知识模态的面向本质计算的差分内容推荐方法,其特征在于,当数据库所提供查询函数为Q3(i,j)和Q4{j,“Value 1”,“Value 2”,...,“Value n”}时,所述步骤S22具体包括:
    判断所查询数据是否为数值型,若为数值型则判断已知数据存储位置信息;
    若已知数据存储位置在用户个人信息数据表中第i行第j列,判断所查询数据属性值是否离散;
    若离散,根据已知属性值信息进行查询,具体包括:若已知第j列所有可能出现的离散属性值,执行Q3(i,j)查询获取数据资源和信息资源;
    若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,根据已知属性值信息进行查询, 具体包括:若已知第j列所有可能出现的离散属性值,则从Q3(a,j)开始逐行进行查询,直到i=b,将查询结果作为新信息资源,若已知具体属性值则执行Q4(j,“Value”)查询,将查询结果与多行记录进行Q3查询的结果相匹配获取数据资源和信息资源;
    若已知数据存储位置为空,则先通过执行Q4{j,“Value 1”,“Value 2”,...,“Value n”}查询不同属性值记录总数,再执行Q3(i,j)查询对结果进行分类以获取数据资源和信息资源。
  8. 根据权利要求2所述的一种跨数据信息知识模态的面向本质计算的差分内容推荐方法,其特征在于,当数据库所提供查询函数为Q3(i,j)和Q4{j,“Value 1”,“Value 2”,...,“Value n”}时,所述步骤S22具体包括:
    判断所查询数据是否为字符型,若为字符型则判断已知数据存储位置信息;
    若已知数据存储位置在用户个人信息数据表中第i行第j列,根据已知属性值信息进行查询,具体包括:若已知第j列共有多个记录的属性值是第i行第j列显示的属性值,则先执行Q4{j,“Value 1”,“Value 2”,...,“Value n”}获得不同属性值的记录总数,再执行Q3(i,j)查询对结果进行分类以获取数据资源和信息资源,若未知第j列所有可能出现的属性值且未知目标用户第i行第j列的具体属性值,则通过逐行执行Q3查询进行模糊分类获取新信息资源;
    若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,根据已知属性值信息进行查询,具体包括:若属性值具有预设选项且未知目标用户具体属性值时,对第a行至第b行的第j列执行Q3(i,j)查询,根据查询结果进行模糊分类,若已知目标用户具体属性值则进一步执行Q4(j,“Value”)查询,并将查询结果与a行至b行的第j列进行Q3(i,j)查询得到的结果进行比对获取新信息资源。
  9. 根据权利要求2所述的一种跨数据信息知识模态的面向本质计算的差分内容推荐方法,其特征在于,当数据库所提供查询函数为Q3(i,j)和Q4{j,“Value 1”,“Value 2”,...,“Value n”}时,所述步骤S22具体包括:
    判断所查询数据是否为代码型,若为代码型则判断已知数据存储位置信息;
    若已知数据存储位置在用户个人信息数据表中第i行第j列,根据已知属性值信息进行查询,具体包括:若属性值存在设定规则且未知具体属性值或未知命名规则,执行Q3(i,j)查询进行模糊分类,若已知目标用户特定属性值则执行Q4{j,“Value 1”,“Value 2”,...,“Value n”}查询分析获取具体属性值;
    若已知数据存储位置在用户个人信息数据表中第i行,i∈[a,b],a、b表示行数且|a-b|≥2,所查询数据在第j列,根据已知属性值信息进行查询,具体包括:若属性值存在设定规则且未知目标用户具体属性值时,对第a行至第b行的第j列执行Q3(i,j)查询,根据查询结果进行模糊分类,若已知目标用户具体属性值,则进一步执行Q4(j,“Value”)查询,将获得结果与执行Q3(i,j)查询结果进行比对,获取新信息资源。
  10. 根据权利要求8或9所述的一种跨数据信息知识模态的面向本质计算的差分内容推荐方法,其特征在于,在根据已知数据存储位置信息和已知属性值信息只能进行模糊分类时,引入与相应数据资源相关的数据资源、信息资源或知识资源进行同模态或跨模态关联融合。
PCT/CN2021/097477 2020-10-31 2021-05-31 跨数据信息知识模态的面向本质计算的差分内容推荐方法 WO2022088674A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011198393.3 2020-10-31
CN202011198393.3A CN112307028B (zh) 2020-10-31 2020-10-31 跨数据信息知识模态的面向本质计算的差分内容推荐方法

Publications (1)

Publication Number Publication Date
WO2022088674A1 true WO2022088674A1 (zh) 2022-05-05

Family

ID=74332303

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/097477 WO2022088674A1 (zh) 2020-10-31 2021-05-31 跨数据信息知识模态的面向本质计算的差分内容推荐方法

Country Status (2)

Country Link
CN (1) CN112307028B (zh)
WO (1) WO2022088674A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112307028B (zh) * 2020-10-31 2021-11-12 海南大学 跨数据信息知识模态的面向本质计算的差分内容推荐方法
CN112527834A (zh) * 2020-12-04 2021-03-19 海南大学 跨模态面向本质计算与推理的内容查询方法及组件

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110064221A1 (en) * 2009-09-11 2011-03-17 Microsoft Corporation Differential privacy preserving recommendation
CN103279499A (zh) * 2013-05-09 2013-09-04 北京信息科技大学 个性化信息检索中用户隐私保护方法
CN109784092A (zh) * 2019-01-23 2019-05-21 北京工业大学 一种基于标签和差分隐私保护的推荐方法
CN109885769A (zh) * 2019-02-22 2019-06-14 内蒙古大学 一种基于差分隐私算法的主动推荐系统及装置
CN110765169A (zh) * 2019-09-06 2020-02-07 平安普惠企业管理有限公司 信息推荐方法、装置、计算机设备及存储介质
CN112307028A (zh) * 2020-10-31 2021-02-02 海南大学 跨数据信息知识模态的面向本质计算的差分内容推荐方法

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101116026B1 (ko) * 2009-12-24 2012-02-13 성균관대학교산학협력단 차이 확률 변수의 원점 모멘트를 이용한 유사성 척도에 기반한 협업 필터링 추천 시스템
US10521824B1 (en) * 2014-01-02 2019-12-31 Outbrain Inc. System and method for personalized content recommendations
CN104050267B (zh) * 2014-06-23 2017-10-03 中国科学院软件研究所 基于关联规则满足用户隐私保护的个性化推荐方法及系统
CN106202331B (zh) * 2016-07-01 2019-08-30 中国传媒大学 分层次隐私保护的推荐系统及基于该推荐系统的作业方法
CN108256000B (zh) * 2017-12-29 2021-06-15 武汉大学 一种基于局部聚类的个性化差分隐私推荐方法
CN110727959A (zh) * 2019-10-15 2020-01-24 南京航空航天大学 一种基于聚类的差分隐私轨迹数据保护方法
CN111556437B (zh) * 2020-05-12 2021-11-16 重庆邮电大学 一种基于差分隐私的个性化位置隐私保护方法
CN111666308B (zh) * 2020-06-03 2022-09-30 国家计算机网络与信息安全管理中心 一种基于行为分析的大数据智能推荐查询方法和系统

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110064221A1 (en) * 2009-09-11 2011-03-17 Microsoft Corporation Differential privacy preserving recommendation
CN103279499A (zh) * 2013-05-09 2013-09-04 北京信息科技大学 个性化信息检索中用户隐私保护方法
CN109784092A (zh) * 2019-01-23 2019-05-21 北京工业大学 一种基于标签和差分隐私保护的推荐方法
CN109885769A (zh) * 2019-02-22 2019-06-14 内蒙古大学 一种基于差分隐私算法的主动推荐系统及装置
CN110765169A (zh) * 2019-09-06 2020-02-07 平安普惠企业管理有限公司 信息推荐方法、装置、计算机设备及存储介质
CN112307028A (zh) * 2020-10-31 2021-02-02 海南大学 跨数据信息知识模态的面向本质计算的差分内容推荐方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
RONGMIN SHAO, LIN ZHANG, RUCHUAN WANG: "Data publishing method based on differential privacy-preserving in D-Pro Per protection framework", JOURNAL OF NANJING UNIVERSITY OF POSTS AND TELECOMMUNICATIONS (NATURAL SCIENCE EDITION), NANJING UNIVERSITY OF POSTS AND TELECOMMUNICATIONS, CN, vol. 36, no. 5, 31 October 2016 (2016-10-31), CN , pages 96 - 104, XP055927278, ISSN: 1673-5439, DOI: 10.14132/j.cnki.1673-5439.2016.05.015 *

Also Published As

Publication number Publication date
CN112307028A (zh) 2021-02-02
CN112307028B (zh) 2021-11-12

Similar Documents

Publication Publication Date Title
US8145582B2 (en) Synthetic events for real time patient analysis
Geng et al. Interestingness measures for data mining: A survey
US20210012028A1 (en) Data product release method or system
US20090287503A1 (en) Analysis of individual and group healthcare data in order to provide real time healthcare recommendations
WO2022088674A1 (zh) 跨数据信息知识模态的面向本质计算的差分内容推荐方法
US20020143577A1 (en) Apparatus and method for prediction and management of subject compliance in clinical research
Wang et al. Generating and evaluating synthetic UK primary care data: preserving data utility & patient privacy
Li et al. Discovering statistically non-redundant subgroups
Kotlar et al. Novel meta-features for automated machine learning model selection in anomaly detection
Fathima et al. Analysis of significant factors for dengue infection prognosis using the random forest classifier
Marín et al. Fuzzy frameworks for mining data associations: fuzzy association rules and beyond
Adday et al. Enhanced vaccine recommender system to prevent COVID-19 based on clustering and classification
Adebayo Predictive model for the classification of hypertension risk using decision trees algorithm
Hart et al. Meeting health care research needs in a kimball integrated data warehouse
De Boeck et al. Dataset anonymization with purpose: A resource allocation use case
Inibhunu et al. State based hidden Markov models for temporal pattern discovery in critical care
Hernandez-Matamoros et al. Comparative Analysis of Local Differential Privacy Schemes in Healthcare Datasets
Giordani et al. Graph data base: an enabling technology for drug prescription patterns analysis
Halvorsen et al. How attacker knowledge affects privacy risks: an analysis using probabilistic programming
Reps et al. Identifying candidate risk factors for prescription drug side effects using causal contrast set mining
Oliveira et al. Towards an intelligent systems to predict nosocomial infections in intensive care
Katsis et al. Assisting discovery in public health
Konijn Detecting interesting differences: Data mining in health insurance data using outlier detection and subgroup discovery
US11894117B1 (en) Discovering context-specific complexity and utilization sequences
Pandey Multimodal event driven N-of-1 analysis of individual lifestyle and health

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21884417

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21884417

Country of ref document: EP

Kind code of ref document: A1