CN103995869B - Data-caching method based on Apriori algorithm - Google Patents

Abstract

A data caching method based on the Apriori algorithm of the present invention establishes a query log for the condition attribute in the disk, calculates the query frequency of each data block, and forms a frequent data block set with multiple data blocks with high query frequency, and calculates the frequent data block The query frequency of condition attributes in the set, multiple condition attributes with high query frequency form a frequent condition attribute set. Use the Apriori algorithm to obtain the frequent condition attribute group set, map the query frequency to the support degree in the Apriori algorithm, obtain the frequent condition attribute group set, cache the data corresponding to the frequent condition attribute group set in the memory, and build an index for the frequent condition attribute . The data caching method can significantly improve data query efficiency in frequent areas, and caching multiple condition attribute groups has higher query efficiency than a single condition attribute, thereby reducing the retrieval pressure of the database and achieving higher query efficiency.

Description

A Data Cache Method Based on Apriori Algorithm

technical field

The invention belongs to the technical field of data query, and in particular relates to a data caching method based on an Apriori algorithm.

Background technique

In recent years, with the rapid development of the Internet, especially the rise of social applications such as Weibo and WeChat, the amount of data has exploded. In 2011, human beings officially entered the ZB era. We have to admit that we are already living in the era of big data. However, big data has been endowed with the characteristics of low value density and various types since its birth, which also determines that massive data will face many problems when querying. When the data size is not too large, the traditional relational database has good performance, high stability, and has been tested by history. But when the amount of data reaches a certain scale, for relational databases, the efficiency is extremely low and unbearable. All in all, relational databases cannot meet the requirements of high concurrent read and write of databases in the era of big data, the requirements for efficient storage and access of massive data, and the requirements for high scalability and high availability of databases.

The discovery of the problem gave birth to a new technology - NoSQL. NoSQL means "Not Just SQL" and is a broad definition of non-relational data storage. It breaks the long-standing unification of relational database and ACID theory. NoSQL data storage does not require a fixed table structure, and usually there is no join operation. In terms of big data access, it has incomparable performance advantages compared to relational databases. However, the current mainstream NoSQL databases mostly use the LIRS algorithm to implement the data caching mechanism. However, the LIRS algorithm cannot effectively count the data that is frequently queried for a long period of time, and cannot adopt a targeted strategy to cache the data to be queried.

Contents of the invention

Aiming at the deficiencies in the prior art, the present invention provides a data caching method based on the Apriori algorithm.

Technical scheme of the present invention is:

A data caching method based on the Apriori algorithm, comprising the following steps:

Step 1: Record conditional attributes in user query statements within T days in the disk in units of days, and create T query logs, that is, user query content.

Step 2: Calculate the query frequency of each data block in the query log, and obtain multiple data blocks with high query frequency according to the obtained data block query frequency to form a frequent data block set.

Step 2.1: Determine the query times of the data in each data block in the T query logs.

Step 2.2: Normalize the number of data queries in each data block: set the recent log ratio to distinguish recent log data from historical log data. When the query times of historical log data in a data block is higher than the upper threshold of historical log data query times , the query times of the historical log data is the upper threshold; when the query times of the recent log data in the data block is higher than the upper threshold of the query times of the recent log data, the query times of the recent log data is The upper threshold.

Step 2.3: Perform weighting operation on the query times of the data in the data blocks after normalization processing: respectively weight and sum the query times of the data in each data block in the T query logs after normalization processing and take the average value to obtain each data How often the block is queried.

Step 2.4: Select multiple data blocks with high query frequency according to the query frequency of each data block, that is, frequent data blocks, and each frequent data block forms a frequent data block set.

Step 3: The condition attributes of each frequent data block form a condition attribute set.

Step 4: Calculate the query frequency of each condition attribute in the condition attribute set, and obtain multiple condition attributes with high query frequency according to the obtained condition attribute query frequency to form a frequent condition attribute set.

Step 4.1: Determine the query times of each condition attribute in the frequent data blocks in the T query logs.

Step 4.2: Normalize the query times of each condition attribute: Distinguish recent log condition attributes and historical log condition attributes according to the ratio of recent logs. The query times of the historical log condition attribute takes the value of the upper threshold; when the query times of the recent log condition attribute is higher than the upper threshold of the query times of the recent log condition attribute, the query times of the recent log condition attribute takes the value of the upper threshold.

Step 4.3: Perform weighting operation on the query times of condition attributes after normalization processing: respectively weight and average the query times of each condition attribute within T days after normalization processing, and obtain the query frequency of each condition attribute.

Step 4.4: According to the obtained query frequency of each condition attribute, select a plurality of condition attributes with high frequency, that is, frequent condition attributes, and each frequent condition attribute forms a frequent condition attribute set.

Step 5: Use the Apriori algorithm and the frequent condition attribute set to obtain the frequent condition attribute group set, the query frequency of the condition attribute is mapped to the support degree in the Apriori algorithm, and the frequent item set obtained by the Apriori algorithm is the frequent condition attribute group set.

Step 6: Cache the data corresponding to the frequent condition attribute group set into the memory, and establish an index for the frequent condition attribute in the frequent condition attribute set, and complete the data cache.

Step 7: When the client needs to query data, perform query operations according to the conditional attributes of the data to be queried: if all the conditional attributes of the data to be queried are frequent conditional attributes of the memory cache, the query result will be obtained directly; If some of the conditional attributes of the queried data are frequent conditional attributes of the memory cache, then query the data in the disk database that meets the conditional attributes in the disk database according to the index of the frequent conditional attributes of the part, and complete the query operation; if the conditional attributes of the data to be queried are not in In the frequent condition attribute set of the memory cache, the data block is loaded from the disk for query operation.

The beneficial effect of the present invention is that a brand-new data caching method is proposed, combined with a NoSQL database, a frequent data column caching area is opened in the memory of the data node, and the data caching method can obviously improve the data query efficiency in the frequent area, while for Data query in other areas will not affect its query operation because no processing is done. Compared with a single condition attribute, caching multiple condition attribute groups has higher query efficiency. The number of condition attributes in a condition attribute group is in the middle Large-scale cache, although part of the cache complete hit rate is sacrificed, this type of cache can better complete the streamlining of intermediate records, reduce the intermediate result set in memory due to partial conditional attribute hits, and quickly index according to frequent conditional attributes Locate the data, thereby reducing the retrieval pressure of the database and achieving higher query efficiency.

Description of drawings

Fig. 1 is operating environment HBase segmentation data table process figure in the specific embodiment of the present invention;

Fig. 2 is an improved data query process diagram in a specific embodiment of the present invention;

Fig. 3 is the flow chart of the data caching method based on Apriori algorithm in the specific embodiment of the present invention;

Fig. 4 is a flow chart of processing different hit situations of query condition attributes in a specific embodiment of the present invention;

Fig. 5 is a comparison diagram of query efficiency of different cache modes in the specific embodiment of the present invention;

FIG. 6 is a comparison diagram of condition attribute hits in different cache modes in a specific embodiment of the present invention.

detailed description

The specific embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings.

In this embodiment, under the Hadoop-HBase environment, the query data and user query behavior are simulated using Sina Weibo user data, and T=7 is used to divide the simulation data into 7 equal parts to simulate query logs at different times.

HBase is a column-oriented NoSQL database that runs on the HDFS file system as part of the Hadoop project. In terms of data reading, HBase adopts a column-based storage method. Compared with a row-based storage method, it reduces redundant data reading during the data reading process, improves data reading efficiency, and makes data retrieval faster and more effective. In terms of storage, HBase divides a large-scale data table into several data areas, that is, data blocks. Each area sequentially stores a certain number of records in the data table. By merging multiple related areas, complete table information can be obtained. . The HBase data table segmentation process is shown in Figure 1.

Areas in HBase correspond to the concept of data blocks. Based on the data caching method of this embodiment, data areas with frequent queries are screened out according to data query conditions, that is, frequent data is fast, and frequent condition attribute data in several areas with the highest frequency are cached in in the memory buffer. When the data in the data area is accessed, different data query operations are performed according to the hit situation between the query condition attribute and the cache in the memory. In the HBase environment, the data query process is shown in Figure 2. The client sends a query request to the data region server, and the data region server returns the query result or further query according to the query situation. If the condition attributes of the data to be queried are all cached in memory Frequent condition attributes, the query result will be obtained directly; if part of the condition attributes of the data to be queried is the frequent condition attributes of the memory cache, then the data in the disk database that meets the condition attributes in the disk database will be queried according to the index of the part of the condition attributes, and the query operation will be completed ; If none of the condition attributes of the data to be queried is in the frequent condition attribute set of the memory cache, load the data block from the disk to perform the query operation. The storage nodes at the Hadoop layer are responsible for loading disk data and performing query operations.

The data caching method based on the Apriori algorithm of the present embodiment is as shown in Figure 3, comprises the following steps:

Step 1: record the conditional attributes in user query statements within T days in the disk in units of days, and create T query logs, that is, user query content.

In this embodiment, the condition attribute is the personal information of the Sina Weibo user's age, gender, location, registration date, and online time. During the implementation process, a total of 7 query logs are created to represent the user query records in the last 7 days.

Step 2: Calculate the query frequency of each data block block in the query log, and obtain multiple data blocks with high query frequency according to the query frequency of the obtained data blocks, that is, the frequent data block set block _fd . Assuming that there are 3 data blocks in total, namely block ₁ , block ₂ , and block ₃ , the query frequency calculation process of data block block ₁ is as follows:

Step 2.1: Determine the query times of each data block data in the 7 query logs.

Obtain the query count Count(t) of block ₁ in 7 days according to the query log count. According to statistics, the query times Count(t) of block ₁ related queries are 1350, 1433, 1236, 1546, 1354, 1029, 1175 when the values of t are 0, 1, 2, 3, 4, 5, and 6 respectively.

Step 2.2: Normalize the number of data queries in each data block: set the recent log ratio to distinguish recent log data from historical log data. When the query times of historical log data in a data block is higher than the upper threshold of historical log data query times When the query times of the historical log data is the upper threshold; when the query times of the recent log data in the data block is higher than the upper threshold of the query times of the recent log data, the query times of the recent log data are set to the upper threshold. upper threshold.

Normalize the query times Count(t) of the data in each data block: set the recent log ratio q _rec to distinguish recent log data from historical log data, q _rec is set according to the actual needs of users, and the value range is 0<q _rec <1 , this embodiment takes q _rec =0.3, then when t<q _rec ×T, that is, the query log data of the previous 5 days belongs to historical log data, and when t≥q _rec ×T, that is, the query log data of the last 2 days belongs to recent For log data, for the historical log data in the data block, set the upper limit threshold Max _his of historical log data query times. Normally, Max _his should be set to 1.5 times the average query times of all records, Max _his = 1400, the data block query When the number of times is higher than the upper limit threshold of historical log data query times, the value of the block data query times is the upper threshold value. For recent log data, set the upper limit threshold of recent log data query times Max _rec , and Max _rec should be set to the average value of all records. 2 times the number of queries, Max _rec = 1700, when the query times of the data block is higher than the upper threshold of recent log data query times, then the query times of the data block is taken as the upper threshold, and the query times are calculated according to the normalized formula (1) Count(t) is normalized:

Since in step 2.1 when t=1 and t=3, the number of related queries has exceeded the upper limit threshold of historical log data queries, so Count(1)=Count(3)=Max _his =1400, and Count _std1 (t) is 1350 ,1400,1236,1400,1354,1029,1175.

By normalizing the number of queries, it is possible to avoid, to a certain extent, the situation that the query frequency of conditional attributes is too high due to the high number of queries in individual days.

Step 2.3: Perform a weighting operation on the number of data block queries Count _std1 (t) after normalization processing: respectively weight and sum the query times of data in each data block in the seven query logs after normalization processing, and take the average value to obtain The data block query frequency FD _block :

Among them, Count _std1 (t) is the query times of data in the data block after normalization processing, and W(t) is a weighting function, which is an increasing function.

In this embodiment, the function part of the monotonically increasing proportional function in the first quadrant is used as the weighting function, that is, W(t)=t+1, where 0≤t≤6

Calculate the frequency of block ₁ :

Step 2.4: Select multiple data blocks with high query frequency according to the query frequency of each data block, that is, frequent data blocks, and each frequent data block forms a frequent data block set, where the query frequency of block ₂ is 5973.13648, and the query frequency of block ₃ is 5973.13648 The frequency is 5294.65. Sort the data blocks by size, and obtain multiple data blocks with high query frequency within 1G after summing up the data blocks sequentially, among which block ₂ belongs to the frequent data block.

Step 3: The conditional attributes of each frequent data block form a conditional attribute set: in this embodiment, the conditional attribute set of the user's age, gender, location, registration date, and online time is a conditional attribute set.

Step 4: Calculate the query frequency of each condition attribute in the condition attribute set, and obtain multiple condition attributes with high query frequency according to the obtained condition attribute query frequency to form a frequent condition attribute set. Taking the age condition attribute as an example, the calculation process of condition attribute query frequency is as follows:

Step 4.1: Determine the number of queries of each condition attribute in the frequent data block in the 7 query logs. In this embodiment, the query related to the age condition attribute is 0, 1, 2, 3, 4, 5, 6 at t The query times are 130, 135, 125, 160, 110, 115, and 120 respectively.

Step 4.2: Normalize the query times of each condition attribute: Distinguish recent log condition attributes and historical log condition attributes according to the ratio of recent logs. The query times of the historical log condition attribute takes the value of the upper threshold; when the query times of the recent log condition attribute is higher than the upper threshold of the query times of the recent log condition attribute, the query times of the recent log condition attribute takes the value of the upper threshold.

Normalize the query times of condition attributes. According to the recent log ratio q _rec = 0.3, the query log condition attributes of the previous 5 days belong to the historical log condition attributes, the query log condition attributes of the last 2 days belong to the recent log condition attributes, and the historical log condition attributes , set the historical log condition attribute query times upper limit threshold Max _his , Max _his =140, when the condition attribute query times is higher than the historical log condition attribute query times upper limit threshold, then the condition attribute query times value is the upper threshold value. For the recent log condition attribute, set the upper limit threshold Max _rec of the recent log condition attribute query times, Max _rec =150, when the condition attribute query times is higher than the recent log condition attribute query times upper threshold value, then the condition attribute query times value is The upper limit threshold, according to the normalization formula (1), carries out normalization processing to the count (t) of conditional attribute query times, because when t=3 obtained in step 4.1, Count (3)=160, has exceeded the historical log query times upper limit threshold , so Count(3)=Max _his =140. Count _std2 (t)Count(stt)d is 130, 135, 125, 140, 110, 115, 120.

Step 4.3: Perform weighting operation on the query times of conditional attributes after normalization processing: respectively weight the query times of each conditional attribute within 7 days after normalization processing, sum and take the average value, that is, obtain the query frequency FD _sa of each conditional attribute :

Also use the function part of the monotonically increasing proportional function in the first quadrant as a weighting function, that is, W(t)=t+1, where 0≤t≤6, calculate the frequency of the age condition attribute:

Step 4.4: According to the obtained query frequency of each condition attribute, select multiple condition attributes with high frequency, that is, frequent condition attributes, and each frequent condition attribute forms a frequent condition attribute set, in which the query frequency of age is 487.8571, and the query frequency of gender is The query frequency is 539.2857143, the query frequency of the region is 632.1428571, the query frequency of registration date is 217.1429, and the query frequency of online time is 103.4923.

Step 5: Use the Apriori algorithm and the frequent condition attribute set to obtain the frequent condition attribute group set, the query frequency of the condition attribute is mapped to the support degree in the Apriori algorithm, and the frequent item set obtained by the Apriori algorithm is the frequent condition attribute group set.

Step 5.1: Let A ₁ =φ, let k be the current highest frequent condition attribute group length, when k=1, it means the frequent condition attribute group set A 1 with length ₁ .

Step 5.2: Count the query frequency of each condition attribute in the frequent condition attribute set. Among them, the frequency corresponding to the condition attributes of age, gender, location, registration date, and online time are 487.8571, 539.2857, 632.1428, 217.1429, and 103.4923 respectively, and the minimum setting is Frequency threshold min _fd = 175, put all condition attributes greater than or equal to the minimum frequency threshold such as age, gender, location, registration date, and online time into A ₁ , and obtain a frequent condition attribute group set A _{1 with a length of 1} .

Step 5.3: Sort the elements in A ₁ according to the conditional attribute names and perform a natural connection to obtain a frequent conditional attribute group candidate set C ₂ with a length of 2, where C ₂ includes the location - registration date, location - age , Location-gender, age-gender, age-registration date, gender-registration date.

Step 5.4: Let A ₂ = φ, query each conditional attribute group in C ₂ , and retrieve all frequent conditional attribute sets, and count the query frequency of each conditional attribute group in C ₂ , where the location - registration date, location - age The frequencies corresponding to attribute groups such as location-gender, age-gender, age-registration date, and gender-registration date are 202.14285, 339.2857, 401.4285, 321.4285, 98.4957, and 135.671 respectively, and the frequency of the conditional attribute group is greater than or equal to the minimum frequency The condition attribute groups such as location-registration date, location-age, location-gender, age-gender and other condition attribute groups of the degree threshold are put into the frequent condition attribute group set A2 whose length is ₂ in A2.

Step 5.5: Sort the elements in A ₂ lexicographically according to the conditional attribute names and perform natural connection to obtain the frequent conditional attribute group candidate set C ₃ with a length of 3, where C ₃ includes the region-gender-age conditional attribute group.

Step 5.6: Let A ₃ = φ, query each conditional attribute group in C ₃ , and retrieve all frequent conditional attribute sets, and count the query frequency of the conditional attribute group in C ₃ , where the conditional attribute group of region-gender-age corresponds to The frequency of each is 183.5714286, and the region-gender-age conditional attribute group whose conditional attribute group frequency is greater than or equal to the minimum frequency threshold is put into the frequent conditional attribute group set A ₃ with a length of 3.

Step 5.7: Sort the elements in A ₃ lexicographically according to the conditional attribute names and perform natural connection to obtain a frequent conditional attribute group candidate set C ₄ with a length of 4, where C ₄ =φ.

Step 5.8: Obtain the frequent query condition attribute group set A of each length in the query log, where A=∪ _k A _k =A ₁ ∪A ₂ ∪…∪A _k :

The frequent condition attributes with a length of 1 include age, gender, location, and registration date, and the corresponding frequencies are 487.8571, 539.2857, 632.1428, and 217.1429, respectively.

The frequent condition attribute groups with a length of 2 include region-registration date, region-age, region-gender, age-gender, and the corresponding frequencies are 202.14285, 339.2857, 401.4285, and 321.4285, respectively.

The frequent condition attribute group with a length of 3 includes region-sex-age, and the frequency is 183.5714286.

Step 6: Cache the data corresponding to the frequent condition attribute group set into the memory, and establish an index for the frequent condition attribute in the frequent condition attribute set, and complete the data cache.

Only 3 columns of conditional attribute data are cached in the memory, and there are 3 sets of cache methods. The first set of memory cache caches age, gender, and region data, and the second set of memory cache caches location-age, region-gender data. The condition attributes of the region are repeated, so no memory space is occupied. The region-gender-age data is cached in the third group of memory caches.

Step 7: When the client needs to query data, perform query operations according to the conditional attributes of the data to be queried: if all the conditional attributes of the data to be queried are frequent conditional attributes of the memory cache, the query result will be obtained directly; If some of the conditional attributes of the queried data are frequent conditional attributes of the memory cache, then query the data in the disk database that meets the conditional attributes in the disk database according to the index of the frequent conditional attributes of the part, and complete the query operation; if the conditional attributes of the data to be queried are not in In the frequent condition attribute set of the memory cache, load the data block from the disk for query operability, that is, if there is a miss, load the data block from the disk for query operation.

For an actual data query, there are three possible different hit situations, as shown in Figure 4.

When the user queries the date of birth condition attribute, the date of birth condition attribute is not cached in the memory, and the condition attribute in the query is not in the memory cache, the data block is loaded from the disk to perform the query operation.

When the user queries the region-date of birth condition attribute group, only a part of the condition attributes in the query is in the memory cache, then query the data in the disk database that meets the region condition attribute according to the index of the region condition attribute, and complete the query operate.

When the user queries region conditional attributes, all the conditional attributes in the query are in the memory cache. At this time, the relevant data can be directly retrieved in the memory and the result returned.

The comparison of average query efficiency under different cache modes is shown in Figure 5. Before applying this method, the average query time of a normal SQLSelect statement is about 1500 milliseconds.

The data caching method of this embodiment can obviously improve the efficiency of data query in the frequent area, but for the data query in other areas, since no processing is done, it will not affect the query operation on it. Cache two and three condition attribute groups have higher query efficiency than a single condition attribute. This is because in the actual query process, the condition query frequency of a single condition attribute is low, and the cache complete hit rate is not ideal. Compared with multi-condition Attribute query, a single-condition attribute cache cannot remove irrelevant records well, and the size of the filtered records is large, which brings huge time overhead to the subsequent index verification work in the database.

The comparison of query hits is shown in Figure 6. Although the complete hit rate of the two-condition attribute group cache is much different than that of the three-condition attribute group cache, its partial hit rate is as high as 63.93%. For the cache with the number of conditional attributes in the conditional attribute group at an intermediate scale, although part of the cache complete hit rate is sacrificed, this type of cache can better complete the streamlining of intermediate records and reduce memory due to partial conditional attribute hits. Intermediate result sets, and quickly locate data according to the frequent condition attribute index, thereby reducing the retrieval pressure of the database and achieving higher query efficiency. The average query efficiency of the two-condition attribute group cache is slightly higher than that of the three-condition attribute group cache. This is the case .

Claims (3)

1. a kind of data cache method based on Apriori algorithm is it is characterised in that comprise the following steps：

Step 1：Record the conditional attribute in user's query statement in T days in disk, set up T inquiry day in units of sky Will, i.e. user's inquiry content；

Step 2：Calculate the inquiry frequent degree of each data block in inquiry log, the size according to obtaining data block inquiry frequent degree obtains The high multiple data blocks of frequent degree must be inquired about, form frequent data item set of blocks；

Step 3：The conditional attribute formation condition community set of each frequent data item block；

Step 4：The inquiry frequent degree of each conditional attribute in design conditions community set, inquires about frequency according to obtaining conditional attribute The size of numerous degree obtains the high multiple conditional attributes of inquiry frequent degree, forms frequent conditional attribute set；

Step 5：Obtain frequent conditional attribute group set using Apriori algorithm and frequent conditional attribute set, conditional attribute Inquiry frequent degree is mapped as the support in Apriori algorithm, and the frequent item set that Apriori algorithm obtains is frequent condition and belongs to Property group set；

Step 6：By corresponding for frequent conditional attribute group set data buffer storage to internal memory, and in frequent conditional attribute set Frequently conditional attribute sets up index, completes data buffer storage；

Step 7：When client needs to carry out data query, according to the conditional attribute of data to be inquired about, carry out inquiry operation： To the frequent conditional attribute of all memory caches of conditional attribute of the data of inquiry, then directly obtain Query Result；To A conditional attribute part for the data of inquiry is the frequent conditional attribute of memory cache, then according to this partly frequent conditional attribute Meet the data of this partial condition attribute in search index disk database, complete inquiry operation；Bar to the data of inquiry Part attribute all not in the frequent conditional attribute set of memory cache, then loads data block from disk and carries out inquiry operation.

2. the data cache method based on Apriori algorithm according to claim 1 is it is characterised in that described step 2 has Body executes as follows：

Step 2.1：Determine the inquiry times of data in each data block in T inquiry log；

Step 2.2：Standardization processing is carried out to the inquiry times of the data in each data block：Arrange recent daily record ratio to distinguish closely Phase daily record data and history log data, when the inquiry times of the history log data in data block are looked into higher than history log data When asking number of times upper limit threshold, then the inquiry times value of this history log data is this upper limit threshold；When recent in data block When the inquiry times of daily record data are higher than recent daily record data inquiry times upper limit threshold, then the inquiry of this recent daily record data time Number value is this upper limit threshold；

Step 2.3：The inquiry times of data in the data block after standardization processing are weighted operating：At respectively to standardization Average after the inquiry times weighted sum of data in each data block in T inquiry log after reason, that is, obtain each data block Inquiry frequent degree；

Step 2.4：The high multiple data blocks of inquiry frequent degree are selected according to the size that each data block inquires about frequent degree, frequently counts According to block, each frequent data item block forms frequent data item set of blocks.

3. the data cache method based on Apriori algorithm according to claim 1 is it is characterised in that described step 4 has Body executes as follows：

Step 4.1：Determine the inquiry times of each conditional attribute in frequent data item block in T inquiry log；

Step 4.2：Standardization processing is carried out to the inquiry times of each conditional attribute：Recent daily record is distinguished according to recent daily record ratio Conditional attribute and history log conditional attribute, when the inquiry times of history log conditional attribute are looked into higher than history log conditional attribute When asking number of times upper limit threshold, then the inquiry times value of this history log conditional attribute is this upper limit threshold；When recent daily record bar When the inquiry times of part attribute are higher than recent Log conditions attribute query number of times upper limit threshold, then this recent Log conditions attribute is looked into Inquiry number of times value is this upper limit threshold；

Step 4.3：Conditional attribute inquiry times after standardization processing are weighted operate：After respectively to standardization processing Sue for peace after the inquiry times weighting of each conditional attribute in T days and average, that is, obtain the inquiry frequent degree of each conditional attribute；

Step 4.4：According to the inquiry frequent degree of each conditional attribute obtaining, select the high multiple conditional attributes of frequent degree, i.e. frequency Numerous conditional attribute, each frequent conditional attribute forms frequent conditional attribute set.