An automatic data hot and cold identification method based on entity objects

Abstract

The application discloses an automatic data cold and hot identification method based on entity objects, relates to the fields of data science and databases, is mainly used in the fields of data storage and calculation, and comprises five parts of entity feature extraction, data temperature initialization, data temperature updating, cold and hot threshold determination and cold and hot data marking, extracts the features of a group of entities contained in data in the dimensions of attributes and behaviors, calculates and updates the data temperature based on the features and feature combinations by using an improved exponential smoothing method and a temperature cooling law, and automatically completes the data cold and hot identification according to dynamic thresholds. The application can more comprehensively utilize entity information of data, adopts an automatic data cold and hot identification method which is more objective, sustainable and quantifiable, can describe the cold and hot degrees of data, characterizes and distinguishes the cold and hot of data from the source, makes the cold and hot identification more efficient, intelligent and reliable, and lays a foundation for further reducing storage and improving calculation performance of database related products.

Description

Technical Field

[0001] This invention relates to the fields of data science and databases, and is mainly used in the fields of data storage and computing. Specifically, it designs an automatic hot and cold data identification method based on entity objects. Background Technology

[0002] With the development of big data and internet technology, the scale of data is growing explosively. The massive amount of data places higher demands on the storage and computing capabilities of databases. Typically, systems need to improve computing performance as much as possible while pursuing storage costs, so that database-related platforms can achieve the effect of "storing less and computing faster", thereby improving the core competitiveness of database-related products.

[0003] The essence of hot and cold data identification lies in distinguishing between hot and cold data. Based on the degree of hotness or coldness, data is stored on different media or using different computational strategies to achieve a balance between storage and computation. Therefore, the identification and evaluation of hot and cold data has become a key research focus in the database field. Traditional hot and cold data identification methods mainly utilize data structure-based methods, such as Least Recently Used (LRU) and Least Recently Used (LFU). However, several problems still exist:

[0004] 1) Directly determining the temperature of data by its relative position or time cannot intuitively quantify the degree of temperature fluctuation, and the temperature changes fluctuate significantly, making it too susceptible to abnormal situations.

[0005] 2) The deeper meaning of the data and the corresponding behaviors behind the data were not taken into account. The features used were too simplistic and did not combine the entity information contained in the data, making it impossible to comprehensively measure the popularity of the data.

[0006] 3) The lack of a data temperature update strategy and the absence of iterative update calculations make it impossible to track temperature changes throughout the entire data lifecycle, thus lacking sustainability.

[0007] 4) There is a lack of a method to automatically identify hot and cold data based on temperature. Summary of the Invention

[0008] The technical problem to be solved by this invention is that the existing methods for identifying hot and cold data mainly suffer from problems such as using only one feature dimension, not utilizing entity information, and having incomplete, unintuitive, and unsustainable descriptions of hot and cold data. In response to the problems in hot and cold data identification, this invention proposes an automatic hot and cold data identification method based on entity objects.

[0009] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0010] An automatic data hot / cold identification method based on entity objects includes five parts: entity feature extraction, data temperature initialization, data temperature update, hot / cold threshold determination, and hot / cold data labeling. The method extracts features of entities in the data in dimensions such as attributes and behaviors. Based on these features, the method combines an improved exponential smoothing method and a temperature cooling law to complete the temperature calculation and update of the data. The method automatically completes the hot / cold identification of the data according to a dynamic threshold.

[0011] Specifically, it includes the following steps;

[0012] Step 1: Identify the entity objects in the data. Each entity object can represent a single entity or a group of entities. Then, extract the attributes and behavioral characteristics of the entity object.

[0013] Step 2: Combining the concept of temperature cooling and Newton's laws, the initial temperature T0 of all data is calculated based on this characteristic using an improved exponential decay method;

[0014] Step 3: Then, the exponential smoothing method is used to calculate the updated temperature T1 of the data, completing the iteration of the data temperature;

[0015] Step 4: Then, based on the temperature distribution characteristics and the Pareto principle, the hot and cold thresholds of the data are automatically determined;

[0016] Step 5, finally achieving automatic hot and cold data identification.

[0017] As a further preferred embodiment of the automatic hot / cold data identification method based on entity objects according to the present invention, the entity feature extraction is specifically as follows:

[0018] Data hot / cold identification based on entity objects: An entity object is defined as one or a group of entities, with a group consisting of two or more entities. Entity features include attribute feature a and behavioral feature b.

[0019] a=Δt=current_time-last_time

[0020] Where Δt represents the time interval between the last occurrence of the entity object in the data generated in the current period and the current time, and its unit can be hour, day, week, or month;

[0021] b represents the frequency γ of the entity object appearing in the current period, and its unit is times.

[0022] As a further preferred embodiment of the automatic hot / cold data identification method based on entity objects according to the present invention, the data temperature initialization is as follows:

[0023] Combining the concept of temperature cooling and Newton's laws, the initial temperature T0 is calculated based on the solid characteristics using an improved exponential decay method. The specific calculation method is as follows:

[0024] T0=λf(a)+(1-λ)f(b)

[0025] Where f(a) and f(b) represent the attribute feature temperature score and behavioral feature temperature score of the entity object, respectively, and λ and (1-λ) represent the contribution of attribute features and behavioral features to the data temperature, respectively.

[0026] The calculation methods for f(a) and f(b) are as follows:

[0027] f(a) = e -αΔt

[0028] f(b) = 2 / (1+e) -αγ )-1

[0029] Where α∈(0,1) represents the temperature decay coefficient, the calculation method for the initial temperature T0 is obtained as follows:

[0030] T0=λe -αΔt +(1-λ)(2 / (1+e -αγ )-1).

[0031] As a further preferred embodiment of the automatic hot / cold data identification method based on entity objects of the present invention, the data temperature update is specifically as follows:

[0032] There are three scenarios for data updates:

[0033] d) Data in T0 also appears in T1, meaning that T0 and T1 share data;

[0034] e) Data in T0 does not appear in T1, meaning data unique to T0;

[0035] f) The missing data in T1 appears in T0, that is, new data is added in T1;

[0036] The above three situations correspond to three different temperature changes, specifically:

[0037] d) Data temperature rises;

[0038] e) Data temperature decay;

[0039] f) Update the temperature of newly added data, i.e. calculate the initial temperature value of newly added data;

[0040] For the three temperature change scenarios mentioned above, the data temperature update methods include two aspects: temperature rise / fall updates and new data updates.

[0041] 3) Methods for updating heating and cooling

[0042] The temperature data is updated using an exponential smoothing method, and the specific calculation method is as follows:

[0043] T1 = α * T0 + (1 - α) * C1

[0044] Where T1 represents the temperature after the current data update, T0 represents the historical temperature of the previous period, α∈(0,1) represents the cooling coefficient of the data, and C1 represents the update status of the data at the current time, i.e., the behavioral characteristics of the entity object, with specific values ​​as follows:

[0045]

[0046] 4) Added a new method for updating temperature data.

[0047] The methods for updating the temperature of newly added data are divided into the following two types:

[0048] a) If the data containing the entity already carries its attributes and behavioral characteristics, use the same calculation method as the initial temperature.

[0049] The method is to calculate using the following formula:

[0050] T I =λe -αΔt +(1-λ)(2 / (1+e -αγ )-1)

[0051] b) If the data containing the entity object does not carry information related to its attributes and behavioral characteristics, or the information is incomplete, then the usual approach should be adopted.

[0052] The numerical method assigns an initial temperature to the data, which is calculated using the following formula:

[0053] T1 = A

[0054] All data have theoretical lower and lower limits for temperature. The midpoint between the upper and lower limits is taken as the initial temperature. For example, if the temperature range is (0,1), 0.5 is taken as the initial temperature value for all data.

[0055] Based on prior knowledge, the boundary temperature threshold between hot and cold data is used as the initial temperature of all data. The hotness or coldness of the data is entirely determined by the impact of subsequent updates on the data.

[0056] If the generation of data is also considered as an iterative update, then the temperature of the new data is equal to the heating constant, that is, A = 1 - α;

[0057] If the current time is the second temperature calculation cycle, that is, the first temperature update, then all T0 in the formula are the temperatures calculated in the data temperature initialization step;

[0058] If the third cycle or all subsequent cycles have been performed, all T0 values ​​in the formula represent the most recent temperature, i.e., the temperature calculated in the previous cycle.

[0059] As a further preferred embodiment of the automatic hot / cold data identification method based on entity objects according to the present invention, the hot / cold threshold determination is specifically as follows:

[0060] Based on the temperature distribution data and the Pareto principle (80 / 20 rule), the core principle for determining the threshold for hot and cold data is T. I If 80% of the data in T1 hits 20% of the data in T0, then T1 must achieve an 80% hit rate in the thermal data of T0. The temperature corresponding to this hit rate is the threshold for hot and cold temperatures. The hit rate (hitting_rate) is calculated as follows:

[0061]

[0062] in, This indicates the amount of data in T1 that also appears in the hot data of T0. The amount of data that appears in T1 and also in T0.

[0063] As a further preferred embodiment of the automatic hot / cold data identification method based on entity objects according to the present invention, the hot / cold data labeling is specifically as follows:

[0064] Based on the calculated hot and cold temperature thresholds, the hot and cold data are labeled. The labeling principle is as follows: data with a temperature greater than or equal to the temperature threshold is labeled as hot data, and data with a temperature less than the temperature threshold is labeled as cold data. Its formal representation is as follows:

[0065]

[0066] The hot_cold_flag indicates whether the data is hot or cold. A value of 1 indicates hot data, while a value of 0 indicates cold data.

[0067] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0068] 1. This invention uses information from multiple dimensions of entity objects to express the meaning behind the data in depth. By utilizing the attributes and behaviors of entities, it provides a more comprehensive description of the temperature of the data.

[0069] 2. This invention proposes a method for calculating the initial temperature of data based on Newton's law of temperature cooling and an improved exponential decay method, which realizes the quantifiability and intuitiveness of the initial temperature.

[0070] 3. This invention proposes a temperature update strategy based on exponential smoothing, which makes the description of data temperature sustainable and the method has universality.

[0071] 4. This invention designs an automatic temperature threshold discrimination mechanism, which reduces human intervention and improves the efficiency and reliability of hot and cold data identification;

[0072] 5. This invention can more comprehensively utilize entity information in data, employing a more objective, sustainable, and quantifiable automatic hot / cold data identification method. It can also differentiate the degree of hotness / coldness, characterizing the data's hotness / coldness from the data source. This makes hot / cold data identification more efficient, intelligent, and reliable, laying the foundation for further reducing storage and improving computing performance in database-related products. Attached Figure Description

[0073] Figure 1 This is a schematic diagram of the module principle of the data hot / cold identification method based on entity objects of the present invention;

[0074] Figure 2 This is a flowchart of the cold / hot data identification method based on entity objects according to the present invention. Detailed Implementation

[0075] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings:

[0076] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0077] The current methods for identifying hot and cold data mainly suffer from problems such as using only one feature dimension, failing to utilize entity information, and providing incomplete, unintuitive, and unsustainable descriptions of hot and cold data.

[0078] This invention addresses the problem of hot and cold data identification by proposing an automatic hot and cold data identification method based on entity objects. It extracts features from entities within the data, including attributes and behaviors. Based on these features, and combining an improved exponential smoothing method and the temperature cooling law, it calculates and updates the data temperature, and automatically identifies hot and cold data according to a dynamic threshold. Specifically, first, the entity objects in the data are identified; each entity object can represent a single entity or a group of entities. Then, the attribute and behavioral features of these entity objects are extracted. Next, combining the concept of temperature cooling and Newton's laws, an improved exponential decay method is used to calculate the initial temperature T0 of all data based on these features. Then, an exponential smoothing method is used to calculate the updated temperature T1 of the data, completing the iteration of data temperature. Finally, based on temperature distribution characteristics and the Pareto principle (80 / 20 rule), the hot and cold data thresholds are automatically determined, ultimately achieving automatic hot and cold data identification.

[0079] This invention primarily utilizes the attributes and behaviors of entities within data as hot and cold features, combining time-based exponential smoothing and entity-based temperature cooling laws to achieve hot and cold data identification. Specifically, this task requires extracting features from entities contained in the data, using these features to calculate the initial and updated temperatures of the data, and then automatically and dynamically distinguishing between hot and cold data based on the meaning of the entities and the temperature distribution.

[0080] The object-based data hot / cold identification method mainly consists of five modules: entity feature extraction, data temperature initialization, data temperature update, hot / cold threshold determination, and hot / cold data labeling. The module structure is as follows: Figure 1 As shown:

[0081] 1. Entity Feature Extraction

[0082] This invention uses entity objects to identify hot and cold data. An entity object is defined as one or a group (two or more) entities, and the entity features include attribute feature a and behavioral feature b.

[0083] a=Δt=current_time-last_time

[0084] Where Δt represents the time interval between the last occurrence of the entity object in the data generated in the current period (last_time) and the present (current_time), and its unit can be hours, days, weeks, months, etc. b represents the frequency γ of the entity object in the current period, and its unit is times.

[0085] 2. Data Temperature Initialization

[0086] In this scheme, all data temperatures have theoretical lower and lower limits. To make the data temperatures more reliable and converge faster, the initial temperature T0 is calculated based on entity characteristics using an improved exponential decay method, combining the concept of temperature cooling and Newton's laws. The specific calculation method is as follows:

[0087] T0=λf(a)+(1-λ)f(b)

[0088] Where f(a) and f(b) represent the attribute feature temperature score and behavioral feature temperature score of the entity object, respectively, and λ and (1-λ) represent the contribution of attribute features and behavioral features to the data temperature, respectively. The calculation methods of f(a) and f(b) are as follows:

[0089] f(a) = e -αΔt

[0090] f(b) = 2 / (1+e) -αγ )-1

[0091] Where α∈(0,1) represents the temperature decay coefficient, therefore, the calculation method for the initial temperature T0 can be obtained:

[0092] T0=λe -αΔt +(1-λ)(2 / (1+e -αγ )-1)

[0093] 3. Data Temperature Update

[0094] There are three possibilities for data updates: a) Data in T0 also appears in T1 (data shared by T0 and T1); b) Data in T0 does not appear in T1 (data unique to T0); c) Data not found in T1 appears in T0 (new data added to T1).

[0095] The above three scenarios correspond to three possible temperature changes: a) data temperature increases; b) data temperature decreases; c) new data temperature is updated (i.e., the initial value of the new data temperature is calculated).

[0096] In response to the three possible temperature changes mentioned above, the data temperature update methods in this section mainly include two aspects: updates during temperature rise and updates during temperature drop, and updates based on newly added data.

[0097] Methods for updating heating and cooling

[0098] Since the warming and cooling of data are relative, this invention primarily employs an exponential smoothing-based method to update the data temperature. The specific calculation method is as follows:

[0099] T1 = α * T0 + (1 - α) * C1

[0100] Where T1 represents the temperature after the current data update, T0 represents the historical temperature of the previous period, α∈(0,1) represents the cooling coefficient of the data, and C1 represents the update status of the data at the current time (i.e., the behavioral characteristics of the entity object), with specific values ​​as follows:

[0101]

[0102] 5) Added a new method for updating temperature data.

[0103] Based on the differences in the information carried by the entity objects, the methods for updating the temperature of newly added data can be divided into the following two types:

[0104] a) If the data containing the entity already carries its attributes and behavioral characteristics, the same measurement method as the initial temperature can be used.

[0105] The calculation method is as follows:

[0106] T I =λe -αΔt +(1-λ)(2 / (1+e -αγ )-1)

[0107] b) If the data containing the entity object does not carry information related to its attributes and behavioral characteristics, or the information is incomplete, an initial temperature can be assigned to the data using the constant method, calculated using the following formula:

[0108] T1 = A

[0109] In this scheme, the temperature of all data has theoretical lower and lower limits. The midpoint between the upper and lower temperature limits can be taken as the initial temperature. For example, if the temperature range is (0,1), 0.5 can be taken as the initial temperature value of all data. Alternatively, the boundary temperature threshold between hot and cold data can be used as the initial temperature of all data based on prior knowledge. The hotness or coldness of the data is entirely determined by the impact of subsequent updates on the data. In addition, if the generation of data is also regarded as an iterative update, the temperature of the new data is equal to the heating constant, that is, A = 1 - α.

[0110] Furthermore, it should be noted that if this is the second temperature calculation cycle, i.e., the first temperature update, then all T0 values ​​in the formula refer to the temperatures calculated during the "data temperature initialization" phase. If the cycle has progressed to the third cycle or all subsequent cycles, then all T0 values ​​in the formula refer to the most recent (i.e., the previous cycle) calculated temperature.

[0111] 4. Determination of hot and cold thresholds

[0112] Based on the temperature distribution data and the Pareto principle (80 / 20 rule), the core principle for determining the threshold for hot and cold data is T. IIf 80% of the data in T1 hits 20% of the data in T0, then the hit rate of T1 in the thermal data of T0 must reach 80%. The temperature corresponding to this hit rate is the threshold for hot and cold temperatures. The hit rate (hitting_rate) is calculated as follows:

[0113]

[0114] in, This indicates the amount of data in T1 that also appears in the hot data of T0. The amount of data that appears in T1 and also in T0.

[0115] 5. Hot and cold data marking

[0116] Based on the calculated hot and cold temperature thresholds, the hot and cold data are labeled. The labeling principle is as follows: data with a temperature greater than or equal to the temperature threshold is labeled as hot data, and data with a temperature less than the temperature threshold is labeled as cold data. Its formal representation is as follows:

[0117]

[0118] The hot_cold_flag indicates whether the data is hot or cold. A value of 1 indicates hot data, while a value of 0 indicates cold data.

[0119] The entity-based data hot / cold identification method mainly consists of five modules: entity feature extraction, data temperature initialization, data temperature update, hot / cold threshold determination, and hot / cold data labeling. Its specific processing flow is as follows: Figure 2 As shown: 1. Entity Feature Extraction

[0120] This invention uses entity objects to identify hot and cold data. An entity object is defined as one or a group (two or more) entities, and the entity features include attribute feature a and behavioral feature b.

[0121] a=Δt=current_time-last_time

[0122] Where Δt represents the time interval between the last occurrence of the entity object in the data generated in the current period (last_time) and the present (current_time), and its unit is days. That is, Δt represents the number of days between the last occurrence of the entity object in the data generated in the current period and the present. b represents the frequency γ of the entity object in the current period, and its unit is times.

[0123] 2. Data Temperature Initialization

[0124] In this scheme, all data temperatures have theoretical lower and lower limits. To make the data temperatures more reliable and converge faster, the initial temperature T0 is calculated based on entity characteristics using an improved exponential decay method, combining the concept of temperature cooling and Newton's laws. The specific calculation method is as follows:

[0125] T0=λf(a)+(1-λ)f(b)

[0126] Where f(a) and f(b) represent the attribute feature temperature score and behavioral feature temperature score of the entity object, respectively. The calculation methods for f(a) and f(b) are as follows:

[0127] f(a) = e -αΔt

[0128] f(b) = 2 / (1+e) -αγ )-1

[0129] Where α = 0.1, it represents the temperature decay coefficient, which is:

[0130] f(a) = e -0.1Δt

[0131] f(b) = 2 / (1+e) -0.1γ )-1

[0132] λ and (1-λ) represent the contributions of attribute features and behavioral features to the data temperature, respectively, with λ taking a value of 0.5, that is:

[0133] T0 = ​​0.5 * f(a) + 0.5 * f(b)

[0134] Therefore, the method for calculating the initial temperature T0 can be obtained:

[0135] T0 = ​​0.5 * e -0.1Δt +0.5*(2 / (1+e -0.1γ )-1)

[0136] 3. Data Temperature Update

[0137] There are three possibilities for data updates: a) Data in T0 also appears in T1 (data shared by T0 and T1); b) Data in T0 does not appear in T1 (data unique to T0); c) Data not found in T1 appears in T0 (new data added to T1).

[0138] The above three scenarios correspond to three possible temperature changes: a) data temperature increases; b) data temperature decreases; c) new data temperature is updated (i.e., the initial value of the new data temperature is calculated).

[0139] In response to the three possible temperature changes mentioned above, the data temperature update methods in this section mainly include two aspects: updates during temperature rise and updates during temperature drop, and updates based on newly added data.

[0140] 1) Methods for updating heating and cooling

[0141] Since the warming and cooling of data are relative, this invention primarily employs an exponential smoothing-based method to update the data temperature. The specific calculation method is as follows:

[0142] T1 = α * T0 + (1 - α) * C1

[0143] Where T1 represents the temperature after the current data update, T0 represents the historical temperature of the previous period, α∈(0,1) represents the cooling coefficient of the data, and C1 represents the update status of the data at the current time (i.e., the behavioral characteristics of the entity object), with specific values ​​as follows:

[0144]

[0145] With α set to 0.1, the temperature update method for this scenario is as follows:

[0146] T1 = 0.1 * T0 + 0.9 * C1

[0147] 2) New data temperature update method

[0148] When the data containing an entity object lacks information about its attributes and behavioral characteristics, or the information is incomplete, an initial temperature can be assigned to the data using a constant method, calculated using the following formula:

[0149] T1 = 1 - α = 0.9

[0150] Furthermore, it should be noted that if this is the second temperature calculation cycle, i.e., the first temperature update, then all T0 values ​​in the formula refer to the temperatures calculated during the "data temperature initialization" phase. If the cycle has progressed to the third cycle or all subsequent cycles, then all T0 values ​​in the formula refer to the most recent (i.e., the previous cycle) calculated temperature.

[0151] 4. Determination of hot and cold thresholds

[0152] Based on the temperature distribution data and the Pareto principle (80 / 20 rule), the core principle for determining the threshold for hot and cold data is T. I If 80% of the data in T1 hits 20% of the data in T0, then the hit rate (hitting_rate) of T1 in the thermal data of T0 must reach 80%. The temperature corresponding to this hit rate is the threshold for hot and cold temperatures. The hit rate (hitting_rate) is calculated as follows:

[0153]

[0154] To facilitate calculations, the temperature range of 0.01-1 was evenly divided into 100 segments during the implementation process, and the above statistics and hit rate calculations were performed for each segment. This indicates the amount of data in T1 that also appears in the hot data of T0. The amount of data that appears in T1 and also in T0.

[0155] 5. Hot and cold data marking

[0156] Based on the calculated hot and cold temperature thresholds, the hot and cold data are labeled. The labeling principle is as follows: data with a temperature greater than or equal to the temperature threshold is labeled as hot data, and data with a temperature less than the temperature threshold is labeled as cold data. Its formal representation is as follows:

[0157]

[0158] Here, `hot_cold_flag` indicates whether the data is hot or cold. The `threshold` value is 0.087, meaning that data with a temperature greater than or equal to 0.087 is considered hot data, and otherwise cold data. Its formal representation is as follows:

[0159]

Claims

1. An automatic hot / cold data identification method based on entity objects, characterized in that: It consists of five parts: entity feature extraction, data temperature initialization, data temperature update, hot and cold threshold determination, and hot and cold data labeling. It extracts features of entities in the data in dimensions such as attributes and behaviors. Based on these features, it combines an improved exponential smoothing method and a temperature cooling law to complete the temperature calculation and update of the data. It automatically completes the hot and cold identification of the data according to the dynamic threshold. Specifically, it includes the following steps; Step 1: Identify the entity objects in the data. Each entity object can represent a single entity or a group of entities. Then, extract the attributes and behavioral characteristics of the entity object. Step 2: Combining the concept of temperature cooling and Newton's laws, the initial temperature T0 of all data is calculated based on this characteristic using an improved exponential decay method; Step 3: Then, the exponential smoothing method is used to calculate the updated temperature T1 of the data, completing the iteration of the data temperature; Step 4: Then, based on the temperature distribution characteristics and the Pareto principle, the hot and cold thresholds of the data are automatically determined; Step 5, ultimately achieving automatic hot / cold data identification; The data temperature initialization is as follows: Combining the concept of temperature cooling and Newton's laws, the initial temperature T0 is calculated based on the solid characteristics using an improved exponential decay method. The specific calculation method is as follows: T0=λf(a)+(1-λ)f(b) Where f(a) and f(b) represent the attribute feature temperature score and behavioral feature temperature score of the entity object, respectively, and λ and (1-λ) represent the contribution of attribute features and behavioral features to the data temperature, respectively. The calculation methods for f(a) and f(b) are as follows: f(a)=e -αΔt f(b)=2 / (1+e -αγ )-1 Where α∈(0,1) represents the temperature decay coefficient, the calculation method for the initial temperature T0 is obtained as follows: T0=λe -αΔt +(1-λ)(2 / (1+e) -αγ )-1).

2. The automatic hot / cold data identification method based on entity objects according to claim 1, characterized in that: The entity feature extraction is as follows: Data hot / cold identification based on entity objects: An entity object is defined as one or a group of entities, with a group consisting of two or more entities. Entity features include attribute feature a and behavioral feature b. a=Δt=current_time-last_time Where Δt represents the time interval between the last occurrence of the entity object in the data generated in the current period and the current time, and its unit can be hour, day, week, or month; b represents the frequency γ of the entity object appearing in the current period, and its unit is times.

3. The automatic hot / cold data identification method based on entity objects according to claim 1, characterized in that: The data temperature update is as follows: There are three scenarios for data updates: a) Data in T0 also appears in T1, meaning that T0 and T1 share data; b) The data in T0 does not appear in T1, that is, the data is unique to T0; c) Data in T1 does not appear in T0, meaning that new data was added in T1; The above three situations correspond to three different temperature changes, specifically: a) Data temperature rises; b) Data temperature decay; c) Update the temperature of newly added data, i.e. calculate the initial temperature value of newly added data; For the three temperature change scenarios mentioned above, the data temperature update methods include two aspects: temperature rise / fall updates and new data updates. 1) Methods for updating heating and cooling The temperature data is updated using an exponential smoothing method, and the specific calculation method is as follows: T1 = α * T0 + (1 - α) * C1 Where T1 represents the temperature after the current data update, T0 represents the historical temperature of the previous period, α∈(0,1) represents the cooling coefficient of the data, and C1 represents the update status of the data at the current time, i.e., the behavioral characteristics of the entity object, with specific values ​​as follows: 2) New data temperature update method The methods for updating the temperature of newly added data are divided into the following two types: a) If the data containing the entity already carries its attributes and behavioral characteristics, use the same calculation method as for the initial temperature. The following formula is used for calculation: T I =λe -αΔt +(1-λ)(2 / (1+e -αγ )-1) b) If the data containing the entity object does not carry information related to its attributes and behavioral characteristics, or the information is incomplete, an initial temperature is assigned to the data using the constant method, i.e., calculated using the following formula: T1 = A All data have theoretical lower and lower limits for temperature. The median of the upper and lower limits is taken as the initial temperature. For example, if the temperature range is (0,1), 0.5 is taken as the initial temperature value for all data. Based on prior knowledge, the boundary temperature threshold between hot and cold data is used as the initial temperature of all data. The hotness or coldness of the data is entirely determined by the impact of subsequent updates on the data. If the generation of data is considered as an iterative update, then the temperature of the new data is equal to the heating constant, which is A = 1 - α; If the current time is the second temperature calculation cycle, that is, the first temperature update, then all T0 in the formula are the temperatures calculated in the data temperature initialization step; If the third cycle or all subsequent cycles have been performed, all T0 values ​​in the formula represent the most recent temperature, i.e., the temperature calculated in the previous cycle.

4. The automatic hot / cold data identification method based on entity objects according to claim 1, characterized in that: The determination of the hot and cold thresholds is as follows: Based on the temperature distribution data and the Pareto principle (80 / 20 rule), the core principle for determining the threshold for hot and cold data is T. I If 80% of the data in T1 hits 20% of the data in T0, then T1 must achieve an 80% hit rate in the thermal data of T0. The temperature corresponding to this hit rate is the threshold for hot and cold temperatures. The hit rate (hitting_rate) is calculated as follows: Where, count(T) 01 ot This indicates the amount of data in T1 that also appears in the hot data of T0. The amount of data that appears in T0 is also present in T0.

5. The automatic hot / cold data identification method based on entity objects according to claim 1, characterized in that: The hot and cold data markings are as follows: Based on the calculated hot and cold temperature thresholds, the hot and cold data are labeled. The labeling principle is as follows: data with a temperature greater than or equal to the temperature threshold is labeled as hot data, and data with a temperature less than the temperature threshold is labeled as cold data. Its formal representation is as follows: The hot_cold_flag indicates whether the data is hot or cold. A value of 1 indicates hot data, while a value of 0 indicates cold data.

Images