Methods, systems, terminal equipment, and media for extracting key factors of click-through rate.

Abstract

This invention discloses a method, system, terminal device, and medium for extracting key factors of click-through rate (CTR), comprising the following steps: determining a first support threshold and a first confidence threshold based on the support and confidence of each itemset in a click scenario, and determining a second support threshold and a second confidence threshold based on the support and confidence of each itemset in a non-click scenario; filtering multiple itemsets in a click scenario according to the first support threshold, the first confidence threshold, and the lift threshold to obtain a first set of items to be filtered, and filtering multiple itemsets in a non-click scenario according to the second support threshold and the second confidence threshold to obtain a second set of items to be filtered; obtaining a candidate set of items to be filtered based on the first and second sets of items to be filtered, performing geodesic filtering on the candidate set to obtain a target set, and extracting key CTR factors from the target set. This invention enables accurate extraction of key CTR factors.

Description

Technical Field

[0001] This invention relates to the field of data analysis, and in particular to a method, system, terminal device, and computer-readable storage medium for extracting key factors of click-through rate. Background Technology

[0002] There are two existing background technologies for assessing the importance of ad click feature metrics: neural network classification data mining algorithms and Apriori association recommendation data mining algorithms.

[0003] Neural network classification algorithms can calculate the weights and impacts of various metrics affecting click-through rates, but they have two major drawbacks:

[0004] (1) Limited layer depth: It can only calculate the impact of indicators on click-through rate, but it is difficult to determine the impact of the algorithm on the component factors of the indicators.

[0005] (2) Limited breadth of indicators: It can only measure the impact of a single indicator, but it is difficult to measure the impact of two or more indicators combined on the click-through rate, which has certain limitations.

[0006] However, the Apriori association recommendation algorithm has the following limitations when dealing with yes-no (i.e., 0-1) type data:

[0007] (1) No standard for threshold confirmation: Although support, confidence and lift can be calculated, there is no specific algorithm for confirming the threshold, nor is there a clear standard. Instead, it relies solely on experience. Furthermore, there is no specific threshold algorithm suitable for yes-no and 0-1 type data such as clicks and purchases, which often results in results that are too general or too strict, and the extracted association combinations do not meet practical requirements.

[0008] (2) Insufficient relative correlation: For 0-1 type data, Apriori considers the correlation itself. In fact, it not only needs to consider the correlation with 1, i.e. click, but also needs to consider the correlation with 0. Because the truly effective correlation requires the overall correlation with click to be greater than the overall correlation with non-click, i.e., it needs to be considered from the perspectives of absolute and relative correlation.

[0009] In summary, existing technologies for screening the importance of ad click feature indicators suffer from incomplete screening scope and inaccurate screening results, resulting in effective click factors that lack specific economic and application value. Summary of the Invention

[0010] The main objective of this invention is to provide a method, system, terminal device, and computer-readable storage medium for extracting key factors of click-through rate (CTR), aiming to achieve accurate extraction of key CTR factors.

[0011] To achieve the above objectives, the present invention provides a method for extracting key factors of click-through rate, the method comprising:

[0012] The first support threshold and the first confidence threshold are determined based on the support and confidence of each itemset in the preset click scenario, and the second support threshold and the second confidence threshold are determined based on the support and confidence of each itemset in the preset non-click scenario.

[0013] The first item set to be filtered is obtained by filtering multiple itemsets in the click scenario based on the first support threshold, the first confidence threshold, and the preset lift threshold; and the second item set to be filtered is obtained by filtering multiple itemsets in the non-click scenario based on the second support threshold and the second confidence threshold.

[0014] Based on the first and second itemsets to be filtered, a candidate set to be filtered is obtained. Geodesic filtering is performed on the candidate set to be filtered to obtain a target itemset, and click-through rate key factors are extracted from the target itemset.

[0015] Optionally, before the steps of determining a first support threshold and a first confidence threshold based on the support and confidence of each itemset in a preset click scenario, and determining a second support threshold and a second confidence threshold based on the support and confidence of each itemset in a preset non-click scenario, the method further includes:

[0016] Obtain preset raw data, and perform correlation statistics based on the raw data to obtain the support, confidence and lift of each item set in the preset click scenario, as well as the support, confidence and lift of each item set in the preset non-click scenario.

[0017] Optionally, the step of determining the first support threshold and the first confidence threshold based on the support and confidence of each itemset under a preset click scenario includes:

[0018] The support of each item set in the click scenario is sorted to obtain the corresponding first median and first upper-level score, and the support corresponding to the first median or the support corresponding to the first upper-level score is determined as the first support threshold.

[0019] The confidence scores of each itemset in the click scenario are sorted to obtain the corresponding second median and second upper-level score, and the confidence score corresponding to the second median or the second upper-level score is determined as the first confidence threshold.

[0020] Optionally, the step of determining the second support threshold and the second confidence threshold based on the support and confidence of each itemset in a preset non-click scenario includes:

[0021] The support scores of each itemset in the non-click scenario are sorted to obtain the corresponding third median and third upper score, and the support score corresponding to the third median or the support score corresponding to the third upper score is determined as the second support threshold.

[0022] The confidence scores of each itemset in the non-click scenario are sorted to obtain the corresponding fourth median and fourth upper-level score, and the confidence score corresponding to the fourth median or the confidence score corresponding to the fourth upper-level score is determined as the second confidence threshold.

[0023] Optionally, the step of filtering multiple itemsets in the click scenario to obtain a first item set to be filtered based on the first support threshold, the first confidence threshold, and a preset lift threshold, and filtering multiple itemsets in the non-click scenario to obtain a second item set to be filtered based on the second support threshold and the second confidence threshold, includes:

[0024] From multiple item sets in the click scenario, select the item set with support greater than the first support threshold, confidence greater than the first confidence threshold, and lift greater than the preset lift threshold as the first item set to be filtered;

[0025] From the multiple item sets in the non-click scenario, select items with support less than the second support threshold and confidence less than the second confidence threshold as the second item set to be filtered.

[0026] Optionally, the step of obtaining a candidate set to be filtered based on the first and second itemsets to be filtered, and then performing geodesic filtering on the candidate sets to be filtered to obtain the target itemset, includes:

[0027] The intersection of the first and second itemsets to be filtered is determined as the candidate itemset to be filtered, and geodesic filtering is performed on the candidate itemset to be filtered according to the preset itemset filtering rules to obtain the corresponding target itemset.

[0028] Optionally, the key factors of click-through rate (CTR) include: a single key factor of CTR, a double key factor of CTR, and a triple key factor of CTR;

[0029] The step of extracting key factors of click-through rate from the target item set includes:

[0030] Extract at least the important single factor, the important dual factor, and the important triple factor of click-through rate from the target item set.

[0031] To achieve the above objectives, the present invention also provides an extraction system for key factors of click-through rate (CTR), the extraction system comprising:

[0032] The determination module is used to determine a first support threshold and a first confidence threshold based on the support and confidence of each itemset in a preset click scenario, and to determine a second support threshold and a second confidence threshold based on the support and confidence of each itemset in a preset non-click scenario.

[0033] The filtering module is used to filter multiple itemsets in the click scenario to obtain a first item set to be filtered based on the first support threshold, the first confidence threshold, and a preset lift threshold, and to filter multiple itemsets in the non-click scenario to obtain a second item set to be filtered based on the second support threshold and the second confidence threshold.

[0034] The extraction module is used to obtain a candidate set to be filtered based on the first item set to be filtered and the second item set to be filtered, perform geodesic filtering on the candidate set to be filtered to obtain a target item set, and extract click-through rate important factors from the target item set.

[0035] In this invention, each functional module of the click-through rate (CTR) important factor extraction system implements the steps of the CTR important factor extraction method described above during operation.

[0036] To achieve the above objectives, the present invention also provides a terminal device, the terminal device comprising: a memory, a processor, and a click-through rate (CTR) important factor extraction program stored in the memory and executable on the processor, wherein the CTR important factor extraction program, when executed by the processor, implements the steps of the CTR important factor extraction method as described above.

[0037] Furthermore, to achieve the above objectives, the present invention also proposes a computer-readable storage medium storing a program for extracting click-through rate (CTR) factors, wherein when the CTR factors extraction program is executed by a processor, it implements the steps of the CTR factors extraction method described above.

[0038] In addition, to achieve the above objectives, the present invention also provides a computer program product comprising a computer program that, when executed by a processor, implements the steps of the method for extracting key factors of click-through rate as described above.

[0039] This invention provides a method, system, terminal device, computer-readable storage medium, and computer program product for extracting key factors of click-through rate (CTR). The method involves determining a first support threshold and a first confidence threshold based on the support and confidence of each itemset in a preset click scenario, and determining a second support threshold and a second confidence threshold based on the support and confidence of each itemset in a preset non-click scenario. A first set of items to be filtered is obtained by filtering multiple itemsets in the click scenario based on the first support threshold, the first confidence threshold, and a preset lift threshold. A second set of items to be filtered is obtained by filtering multiple itemsets in the non-click scenario based on the second support threshold and the second confidence threshold. A candidate set of items to be filtered is obtained based on the first and second sets of items to be filtered. Geodesic filtering is performed on the candidate set of items to be filtered to obtain a target set, and key CTR factors are extracted from the target set.

[0040] Compared to existing neural network classification data mining algorithms and Apriori association recommendation data mining algorithms, this invention overcomes the limitations of these algorithms. In terms of depth, it effectively reaches the indicator factor layer, rather than the indicator layer. In terms of breadth, it can not only measure the impact of a single indicator on click results but also calculate the impact of multiple indicator combinations on click-through rates, thus compensating for the limitations of neural network classification in terms of breadth and depth. Furthermore, this invention proposes a threshold optimization algorithm for support and confidence in association scenarios related to click-through rates, overcoming the limitation of the Apriori algorithm lacking a threshold standard algorithm in this field. It also includes a geodesic filter, making the extraction results more accurate and effective. In summary, this technology, by combining the strengths of various algorithms, possesses significant economic and application value in efficiently identifying effective screening factors for ad clicks. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the hardware operating environment involved in the embodiments of the present invention;

[0042] Figure 2 This is a schematic diagram of the process involved in an embodiment of the method for extracting key factors of click-through rate of the present invention;

[0043] Figure 3 This is a schematic diagram of the original data involved in an embodiment of the method for extracting key factors of click-through rate of the present invention;

[0044] Figure 4 This is a schematic diagram of the discretized original data involved in an embodiment of the method for extracting key factors of click-through rate of the present invention;

[0045] Figure 5This is a first schematic diagram of a three-dimensional itemset involved in an embodiment of the method for extracting key factors of click-through rate of the present invention;

[0046] Figure 6 This is a second schematic diagram of a three-dimensional itemset involved in an embodiment of the method for extracting key factors of click-through rate of the present invention;

[0047] Figure 7 This is a schematic diagram illustrating the screening principle involved in an embodiment of the method for extracting key factors of click-through rate according to the present invention.

[0048] Figure 8 This is a schematic diagram of the screening results involved in an embodiment of the method for extracting key factors of click-through rate of the present invention;

[0049] Figure 9 This is a schematic diagram of the geodesic filtering results involved in an embodiment of the method for extracting key factors of click-through rate of the present invention.

[0050] Figure 10 This is a schematic diagram of the functional modules of an embodiment of the click-through rate key factor extraction system of the present invention.

[0051] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0052] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0053] like Figure 1 As shown, Figure 1 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of the present invention.

[0054] It should be noted that the terminal device in the embodiments of the present invention can be a device for extracting important factors of click-through rate, and the terminal device can specifically be a smartphone, personal computer, server, etc.

[0055] like Figure 1As shown, the device may include: a processor 1001, such as a CPU; a network interface 1004; a user interface 1003; a memory 1005; and a communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0056] Those skilled in the art will understand that Figure 1 The device structure shown does not constitute a limitation on the device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0057] like Figure 1 As shown, the memory 1005, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a program for extracting key factors of click-through rate (CTR). The operating system is a program that manages and controls the hardware and software resources of the device, supporting the operation of the CTR key factor extraction program and other software or programs. Figure 1 In the device shown, the user interface 1003 is mainly used for data communication with the client; the network interface 1004 is mainly used for establishing a communication connection with the server; and the processor 1001 can be used to call the extraction program for click-through rate key factors stored in the memory 1005 and perform the following operations:

[0058] The first support threshold and the first confidence threshold are determined based on the support and confidence of each itemset in the preset click scenario, and the second support threshold and the second confidence threshold are determined based on the support and confidence of each itemset in the preset non-click scenario.

[0059] The first item set to be filtered is obtained by filtering multiple itemsets in the click scenario based on the first support threshold, the first confidence threshold, and the preset lift threshold; and the second item set to be filtered is obtained by filtering multiple itemsets in the non-click scenario based on the second support threshold and the second confidence threshold.

[0060] Based on the first and second itemsets to be filtered, a candidate set to be filtered is obtained. Geodesic filtering is performed on the candidate set to be filtered to obtain a target itemset, and click-through rate key factors are extracted from the target itemset.

[0061] Furthermore, before the steps of determining the first support threshold and the first confidence threshold based on the support and confidence of each itemset in a preset click scenario, and determining the second support threshold and the second confidence threshold based on the support and confidence of each itemset in a preset non-click scenario, the processor 1001 can also call the click rate important factor extraction program stored in the memory 1005, and further perform the following operations:

[0062] Obtain preset raw data, and perform correlation statistics based on the raw data to obtain the support, confidence and lift of each item set in the preset click scenario, as well as the support, confidence and lift of each item set in the preset non-click scenario.

[0063] Furthermore, the processor 1001 can also be used to call the click-through rate key factor extraction program stored in the memory 1005 to perform the following operations:

[0064] The support of each item set in the click scenario is sorted to obtain the corresponding first median and first upper-level score, and the support corresponding to the first median or the support corresponding to the first upper-level score is determined as the first support threshold.

[0065] The confidence scores of each itemset in the click scenario are sorted to obtain the corresponding second median and second upper-level score, and the confidence score corresponding to the second median or the second upper-level score is determined as the first confidence threshold.

[0066] Furthermore, the processor 1001 can also be used to call the click-through rate key factor extraction program stored in the memory 1005 to perform the following operations:

[0067] The support scores of each itemset in the non-click scenario are sorted to obtain the corresponding third median and third upper score, and the support score corresponding to the third median or the support score corresponding to the third upper score is determined as the second support threshold.

[0068] The confidence scores of each itemset in the non-click scenario are sorted to obtain the corresponding fourth median and fourth upper-level score, and the confidence score corresponding to the fourth median or the confidence score corresponding to the fourth upper-level score is determined as the second confidence threshold.

[0069] Furthermore, the processor 1001 can also be used to call the click-through rate key factor extraction program stored in the memory 1005 to perform the following operations:

[0070] From multiple item sets in the click scenario, select the item set with support greater than the first support threshold, confidence greater than the first confidence threshold, and lift greater than the preset lift threshold as the first item set to be filtered;

[0071] From the multiple item sets in the non-click scenario, select the item sets with support less than the second support threshold and confidence less than the second confidence threshold as the second item set to be filtered.

[0072] Furthermore, the processor 1001 can also be used to call the click-through rate key factor extraction program stored in the memory 1005, and also perform the following operations:

[0073] The intersection of the first and second itemsets to be filtered is determined as the candidate itemset to be filtered, and geodesic filtering is performed on the candidate itemset to be filtered according to the preset itemset filtering rules to obtain the corresponding target itemset.

[0074] Furthermore, the key factors of click-through rate (CTR) include: single CTR factor, two CTR factors, and three CTR factors;

[0075] Processor 1001 can also be used to call the click-through rate key factor extraction program stored in memory 1005, and also perform the following operations:

[0076] At least the important single factor, the important dual factor, and the important triple factor of click-through rate are extracted from the multiple target item sets.

[0077] Reference Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the method for extracting key factors of click-through rate according to the present invention.

[0078] This embodiment provides an example of a method for extracting key factors of click-through rate. It should be noted that although the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0079] Step a: Obtain preset raw data, and perform correlation statistics based on the raw data to obtain the support, confidence and lift of multiple itemsets in preset click scenarios, as well as the support, confidence and lift of multiple itemsets in preset non-click scenarios.

[0080] It should be noted that in this embodiment, before extracting the key factors influencing click-through rate (CTR), it is necessary to obtain raw data in advance. This raw data contains various click factors, and each click factor is discretized and correlated statistically analyzed. For example... Figure 3The diagram shows the original data. The original data includes value parameters, method parameters, gender parameters, income parameters, and age parameters for two scenarios: "0 clicks" and "1 click." In this embodiment, "BUY=1" represents "1 click," meaning a click action occurred (click scenario). Similarly, "BUY=0" represents "0 clicks," meaning no click action occurred (non-click scenario). After acquiring a large amount of data containing value, method, gender, income, and age parameters, the terminal device will further discretize these parameters.

[0081] Specifically, for example, such as Figure 4 The diagram shown illustrates the original data after discretization. Value, income, and age are discretized. Taking value x as an example, max is the maximum value and min is the minimum value. First, the data is divided into four intervals based on max and min, with interval m defined as follows:

[0082]

[0083] The four intervals include:

[0084] [min,m), [m,2m), [2m,3m), [3m,max)

[0085] At this point, if x∈[min,m), then x is represented as V1; if x∈[m,2m), then x is represented as V2; if x∈[2m,3m), then x is represented as V3; if x∈[3m,max), then x is represented as V4. The original data after discretization is as follows: Figure 4 As shown, the value level, method, sex, income level, and age level are as follows:

[0086] value-Level = V1, V2, V3, V4

[0087] Pmethod = CARD, CREDIT, CASH

[0088] Sex = M, F

[0089] income-Level = I1, I2, I3, I4

[0090] Age-level = A1, A2, A3, A4

[0091] The discretized raw data above is then analyzed for correlation, with the support, confidence, and lift calculated for buy=1 and buy=0 respectively. An itemset is a set of items. An itemset containing K items is called a K-itemset. For example, {X,Y}={value-Level=V1,BUY=1} is a 2-itemset, and {value-Level=V1pmethod=CASH,SEX=M,income-Level=I3,BUY=1,age-Level=A3} is a 6-itemset. In a K-itemset, the number of items is at least 2, and it must contain at least one item named BUY. Y in the itemset is defined as BUY.

[0092] Furthermore, the support of each itemset in all itemsets is calculated. Taking a two-itemset {X,Y} as an example, the probability that X and Y in the itemset occur simultaneously is called the support of the association rule (also known as the relative support).

[0093]

[0094] Calculate the confidence level of each itemset in all itemsets; that is, the probability that itemset Y occurs if itemset X occurs is the confidence level of the association rule.

[0095]

[0096] Where σ represents the frequency of a given itemset, and N represents the total number of items.

[0097] Calculate the lift of each itemset in all itemsets, which is the ratio of the probability that Y is contained in the itemset given X to the probability that Y is contained in the itemset without X. Generally, the lift should be greater than 1; the higher the lift, the stronger the association.

[0098]

[0099] In this embodiment, it is necessary to obtain itemsets containing Y as much as possible. Ultimately, as... Figure 5 The first schematic diagram of the three-dimensional itemset shows the items with relevance when BUY=1, filtered by minimum reference values ​​of support 0.01, confidence 0.01, and support 0.01. For example... Figure 6 The second schematic diagram of the three-dimensional itemset shows the items with relevance when BUY=0, obtained by filtering with support of 0.01, confidence of 0.01, and support of 0.01 as the minimum reference values.

[0100] Furthermore, the method for extracting key factors of click-through rate in this invention includes:

[0101] Step S10: Determine the first support threshold and the first confidence threshold based on the support and confidence of each itemset in the preset click scenario, and determine the second support threshold and the second confidence threshold based on the support and confidence of each itemset in the preset non-click scenario.

[0102] After obtaining the support and confidence scores for BUY=1 and BUY=0, the terminal device will further determine the support and confidence thresholds for BUY=1 and BUY=0 based on multiple support scores and multiple thresholds.

[0103] Furthermore, in step S10 above, "determining the first support threshold and the first confidence threshold based on the support and confidence of each itemset under a preset click scenario" may include:

[0104] Step S101: Sort the support of each item set in the click scenario to obtain the corresponding first median and first upper score, and determine the support corresponding to the first median or the support corresponding to the first upper score as the first support threshold.

[0105] Step S102: Sort the confidence scores of each itemset in the click scenario to obtain the corresponding second median and second upper-level score, and determine the confidence score corresponding to the second median or the confidence score corresponding to the second upper-level score as the first confidence threshold.

[0106] It should be noted that, in this embodiment, the first threshold includes a first support threshold and a first confidence threshold. Thresholds are calculated for multiple confidence and support levels for BUY=1 using either the median or upper quantile. In practical applications, if the screening criteria are strict, the upper quantile can be used as the threshold; if the screening criteria are slightly broader, the median can be used as the threshold.

[0107] Specifically, for example, taking Support as an example, if Support has a total of n values, and the Support values ​​are arranged in order from the largest to the smallest, the median M (approximately 1 / 2 of n) is calculated based on the sorted n Support values:

[0108]

[0109] And calculate the corresponding superordinate score T (approximately 1 / 2 of M) based on the sorted n Support values:

[0110]

[0111] Furthermore, the threshold for Support is set as the Support value corresponding to the upper quantile or the median, and the threshold for Confidence is set as the Confidence value corresponding to the upper quantile or the median. The median is larger than the upper quantile; the median corresponds to a wider range, and the upper quantile corresponds to a narrower range.

[0112] Furthermore, in step S10 above, "determining the second support threshold and the second confidence threshold based on the support and confidence of each itemset in a preset non-click scenario" may include:

[0113] Step S103: Sort the support of each itemset in the non-click scenario to obtain the corresponding third median and third upper score, and determine the support corresponding to the third median or the support corresponding to the third upper score as the second support threshold.

[0114] Step S104: Sort the confidence scores of each itemset in the non-click scenario to obtain the corresponding fourth median and fourth upper-level score, and determine the confidence score corresponding to the fourth median or the confidence score corresponding to the fourth upper-level score as the second confidence threshold.

[0115] It should be noted that, in this embodiment, the second threshold includes: a second support threshold and a second confidence threshold. Similarly, the thresholds for multiple confidence and support values ​​of BUY=0 can be solved using a method based on the median or upper-level score.

[0116] Specifically, for example, multiple Support values ​​when BUY=0 are arranged in order from largest to smallest. The median M and the uppermost score T are calculated based on the sorted Support values. The Support value corresponding to the median M or the Support value corresponding to the uppermost score T is then used to determine the support threshold when BUY=0. At the same time, multiple Confidence values ​​are arranged in order from largest to smallest. The median M and the uppermost score T are calculated based on the sorted Confidence values. The Confidence value corresponding to the median M or the Confidence value corresponding to the uppermost score T is then used to determine the confidence threshold when BUY=0.

[0117] Furthermore, the method for extracting key factors of click-through rate in this invention also includes:

[0118] Step S20: Filter multiple itemsets in the click scenario according to the first support threshold, the first confidence threshold, and a preset lift threshold to obtain a first item set to be filtered; and filter multiple itemsets in the non-click scenario according to the second support threshold and the second confidence threshold to obtain a second item set to be filtered.

[0119] It should be noted that in this embodiment, the preset lift threshold is 1, and the lift corresponding to the first set of items to be filtered needs to be greater than 1.

[0120] The terminal device filters multiple itemsets under BUY=1 according to the first support threshold, the first confidence threshold and the preset lift threshold 1 to obtain the corresponding itemsets to be filtered, and filters multiple itemsets under BUY=0 according to the second support threshold and the second confidence threshold to obtain the corresponding itemsets to be filtered.

[0121] Further, in step S20 above, "filtering multiple itemsets in the click scenario according to the first support threshold, the first confidence threshold, and a preset lift threshold to obtain a first itemset to be filtered, and filtering multiple itemsets in the non-click scenario according to the second support threshold and the second confidence threshold to obtain a second itemset to be filtered," may include:

[0122] Step S201: Select the first set of items to be filtered from multiple item sets in the click scenario, where the support is greater than the first support threshold, the confidence is greater than the first confidence threshold, and the lift is greater than the preset lift threshold.

[0123] Step S202: Select items from multiple itemsets in the non-click scenario that have support less than the second support threshold and confidence less than the second confidence threshold as the second itemset to be filtered.

[0124] It should be noted that, in this embodiment, the support of the first set of items to be filtered must be greater than the support threshold, while the support of the second set of items to be filtered must be less than the first support threshold; the confidence of the first set of items to be filtered must be greater than the first confidence threshold, while the confidence of the second set of items to be filtered must be less than the second confidence threshold; and the lift of the first set of items to be filtered must be greater than 1.

[0125] Specifically, for example, such as Figure 7 The diagram showing the screening principle, and as shown Figure 8 The diagram showing the filtering results illustrates that after sorting multiple support and confidence scores under BUY=1 to obtain the corresponding first support threshold and first confidence threshold, the terminal device needs to filter out itemsets that are greater than both the first support threshold and the first confidence threshold as the itemsets to be filtered under BUY=1. Similarly, after sorting multiple support and confidence scores under BUY=0 to obtain the corresponding second support threshold and second confidence threshold, the terminal device needs to filter out itemsets that are less than both the second support threshold and the second confidence threshold as the itemsets to be filtered under BUY=0.

[0126] Step S30: Based on the first item set to be filtered and the second item set to be filtered, a candidate set to be filtered is obtained. Geodesic filtering is performed on the candidate set to be filtered to obtain a target item set. Click-through rate important factors are extracted from the target item set.

[0127] It should be noted that, in this embodiment, the geodesic filtering method includes: determining whether the degree of BUY=1 (Support, Confidence) is higher than the degree of BUY=0, and the difference between the two is the geodesic, thereby filtering out the itemsets corresponding to the geodesic greater than 0 as the target itemsets.

[0128] After the terminal device filters out multiple itemsets to be filtered under BUY=1 based on support threshold, confidence threshold and lift, in order to make the results more accurate, it further performs geodesic filtering on the multiple itemsets to be filtered to obtain more accurate target itemsets, and then extracts factors that have an important impact on effective click behavior from multiple target itemsets.

[0129] Further, in step S30 above, "obtaining a candidate set to be filtered based on the first and second candidate sets to be filtered, performing geodesic filtering on the candidate sets to be filtered to obtain a target set, and extracting click-through rate key factors from the target set" may include:

[0130] Step S301: The intersection of the first item set to be filtered and the second item set to be filtered is determined as the candidate set to be filtered, and geodesic filtering is performed on the candidate set to be filtered according to the preset item set filtering rules to obtain the corresponding target item set.

[0131] It should be noted that, in this embodiment, before performing geodesic filtering, the intersection of the first and second itemsets to be filtered needs to be obtained as a candidate itemset, and geodesic filtering is then performed on this candidate itemset. The itemset filtering rules include: only itemsets with a degree higher than that with a degree higher than that with a degree higher than that with a degree higher than that with a degree higher than that with a degree higher than that with a degree lower ...

[0132] T lhs =support(buy=1)-support(buy=0)+confidence(buy=1)-confidence(buy=0)

[0133] Among them, T 1hs This represents the distance between geodesic lines.

[0134] Specifically, for example, taking the candidate set to be filtered {age-Level=A1,BUY=1} as an example, the corresponding geodesic distance is:

[0135] T 1hs =Support{age-Level=A1,BUY=1}-Support{age-Level=A1,BUY=0}+Confidence{age-Level=A1,BUY=1}-Confidence{age-Level=A1,BUY=0}

[0136] Furthermore, such as Figure 9 The diagram showing the geodesic filtering results indicates that although the Lift values ​​of the itemsets corresponding to [5]* and [6]* are positive (itemsets with positive Lift are retained in the Apriori association recommendation algorithm), in this invention, T 1hs Itemsets with negative geodesic distances are useless and need to be removed.

[0137] In this embodiment, after obtaining the support and confidence scores for BUY=1 and BUY=0, the terminal device further selects support and confidence thresholds for BUY=1 and BUY=0 based on multiple support and multiple threshold values. The terminal device filters multiple itemsets for BUY=1 based on a first support threshold, a first confidence threshold, and a preset lift threshold 1 to obtain corresponding itemsets to be filtered. It then filters multiple itemsets for BUY=0 based on a second support threshold and a second confidence threshold to obtain corresponding itemsets to be filtered. After selecting multiple itemsets to be filtered for BUY=1 based on the support threshold, confidence threshold, and lift, the terminal device further obtains corresponding candidate sets to be filtered. To make the results more accurate, geodesic filtering is performed on these candidate sets to obtain more accurate target itemsets, thereby extracting factors that significantly influence effective click behavior from multiple target itemsets.

[0138] Compared to existing neural network classification data mining algorithms and Apriori association recommendation data mining algorithms, this invention overcomes the limitations of these algorithms. In terms of depth, it effectively reaches the indicator factor layer, rather than just the indicator layer; in terms of breadth, it can not only measure the impact of a single indicator on click results but also calculate the impact of multiple indicator combinations on 0-1 type click effects, thus compensating for the limitations of neural network classification in terms of breadth and depth. Furthermore, this invention proposes a threshold optimization algorithm for support and confidence of 0-1 type associations, overcoming the limitation of the Apriori algorithm lacking a threshold standard algorithm in this field. It also specifically adds geodesic filtering for 0-1 type associations, extracting both absolute and relative results, making the results more accurate and effective. In summary, this technology, by combining the strengths of various algorithms, possesses significant economic and application value in efficiently identifying effective screening factors for ad clicks.

[0139] Furthermore, based on the first embodiment of the method for extracting key factors of click-through rate of the present invention described above, a second embodiment of the method for extracting key factors of click-through rate of the present invention is proposed.

[0140] The main difference between this embodiment and the first embodiment described above is that, in step S30, "extracting key factors of click-through rate from the target item set" may include:

[0141] Step S302: Extract at least the important single factor, the important double factor, and the important triple factor of click-through rate from the target item set.

[0142] It should be noted that, in this embodiment, the important factors of click-through rate (CTR) may include: single important CTR factor, double important CTR factor, and triple important CTR factor, etc. In this embodiment, the type of important CTR factor is not specifically limited. In addition to the single factor, double factor, and triple factor mentioned above, more combinations of important CTR factors may be included.

[0143] Specifically, for example, such as Figure 9 The diagram shown illustrates the geodesic filtering results. The terminal device extracts and filters multiple candidate sets based on geodesic distances to obtain multiple target item sets. Then, it obtains corresponding click-through rate (CTR) factors from these target item sets. In this embodiment, the CTR factors may include:

[0144] Single factor: {age.Level = A1};

[0145] Two-factor: {pmethod=CARD,age.Level=A1}, {sex=M,age.Level=A1}, and {sex=F,age.Level=A1};

[0146] Three factors: {pmethod=CARD,sex=F,age.Level=A2};

[0147] As can be seen, a total of 5 key factors of click-through rate were extracted. This invention makes up for the deficiency of neural network classification algorithms that can only rely on a single indicator, and also makes up for the limitation of association algorithms in judging 0-1.

[0148] Furthermore, embodiments of the present invention also propose an extraction system for key factors of click-through rate, referring to... Figure 4 , Figure 4 This is a functional module diagram of an embodiment of the click-through rate (CTR) key factor extraction system of the present invention. Figure 4 As shown, the extraction system for key factors of click-through rate of the present invention includes:

[0149] The determination module 10 is used to determine a first support threshold and a first confidence threshold based on the support and confidence of each itemset in a preset click scenario, and to determine a second support threshold and a second confidence threshold based on the support and confidence of each itemset in a preset non-click scenario.

[0150] The filtering module 20 is used to filter multiple itemsets in the click scenario to obtain a first itemset to be filtered based on the first support threshold, the first confidence threshold, and a preset lift threshold, and to filter multiple itemsets in the non-click scenario to obtain a second itemset to be filtered based on the second support threshold and the second confidence threshold.

[0151] The extraction module 30 is used to obtain a candidate set to be filtered based on the first item set to be filtered and the second item set to be filtered, perform geodesic filtering on the candidate set to be filtered to obtain a target item set, and extract click-through rate important factors from the target item set.

[0152] Furthermore, the extraction system for key factors of click-through rate of the present invention also includes:

[0153] The acquisition module is used to acquire preset raw data, and perform correlation statistics based on the raw data to obtain the support, confidence and lift of multiple itemsets in preset click scenarios, as well as the support, confidence and lift of multiple itemsets in preset non-click scenarios.

[0154] Further, module 10 is defined as including:

[0155] The first determining unit is used to sort the support of each item set in the click scenario to obtain the corresponding first median and first upper-level score, and to determine the support corresponding to the first median or the support corresponding to the first upper-level score as the first support threshold.

[0156] The second determining unit is used to sort the confidence scores of each itemset in the click scenario to obtain the corresponding second median and second upper-level score, and to determine the confidence score corresponding to the second median or the confidence score corresponding to the second upper-level score as the first confidence threshold.

[0157] Furthermore, the determining module 10 also includes:

[0158] The third determining unit is used to sort the support of each itemset in the non-click scenario to obtain the corresponding third median and third upper score, and to determine the support corresponding to the third median or the support corresponding to the third upper score as the second support threshold.

[0159] The fourth determining unit is used to sort the confidence scores of each itemset in the non-click scenario to obtain the corresponding fourth median and fourth upper-level score, and to determine the confidence score corresponding to the fourth median or the confidence score corresponding to the fourth upper-level score as the second confidence threshold.

[0160] Furthermore, the filtering module 20 includes:

[0161] The first filtering unit is used to filter out the first set of items to be filtered from multiple item sets under the click scenario, where the support is greater than the first support threshold, the confidence is greater than the first confidence threshold, and the lift is greater than the preset lift threshold.

[0162] The second filtering unit is used to filter out itemsets with support less than the second support threshold and confidence less than the second confidence threshold from multiple itemsets in the non-click scenario as the second itemset to be filtered.

[0163] Furthermore, the extraction module 30 includes:

[0164] The filtering unit is used to determine the intersection of the first item set to be filtered and the second item set to be filtered as the candidate set to be filtered, and to perform geodesic filtering on the candidate set to be filtered according to the preset item set filtering rules to obtain the corresponding target item set.

[0165] Furthermore, the extraction module 30 also includes:

[0166] An extraction unit is used to extract at least the click-through rate (CTR) important single factor, the CTR important dual factor, and the CTR important triple factor from the target item set.

[0167] The specific implementation methods of each functional module of the click-through rate (CTR) important factor extraction system of the present invention are basically the same as those of the above-described CTR important factor extraction method embodiments, and will not be repeated here.

[0168] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a program for extracting click-through rate (CTR) important factors. When the CTR important factor extraction program is executed by a processor, it implements the steps of the CTR important factor extraction method as described above.

[0169] The various embodiments of the click-through rate (CTR) important factor extraction system and computer-readable storage medium of the present invention can be referred to the various embodiments of the click-through rate (CTR) important factor extraction method of the present invention, and will not be repeated here.

[0170] Furthermore, embodiments of the present invention also provide a computer program product, the computer program product comprising a computer program, which, when executed by a processor, implements the steps of the click-through rate (CTR) important factor extraction method as described in any of the embodiments of the above-described CTR important factor extraction method.

[0171] The specific embodiments of the computer program product of the present invention are basically the same as the embodiments of the above-mentioned method for extracting important factors of click-through rate, and will not be described in detail here.

[0172] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0173] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0174] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a smartphone, personal computer, server, etc.) to execute the methods described in the various embodiments of the present invention.

[0175] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for extracting key factors of click-through rate, characterized in that, The method for extracting key factors of click-through rate includes the following steps: The first support threshold and the first confidence threshold are determined based on the support and confidence of each itemset in the preset click scenario, and the second support threshold and the second confidence threshold are determined based on the support and confidence of each itemset in the preset non-click scenario. The first item set to be filtered is obtained by filtering multiple itemsets in the click scenario based on the first support threshold, the first confidence threshold, and the preset lift threshold; and the second item set to be filtered is obtained by filtering multiple itemsets in the non-click scenario based on the second support threshold and the second confidence threshold. Based on the first and second itemsets to be filtered, a candidate set to be filtered is obtained. Geodesic filtering is performed on the candidate set to be filtered to obtain a target itemset, and click-through rate important factors are extracted from the target itemset. The geodesic filtering method includes: determining whether the support and confidence of BUY=1 are higher than the support and confidence of BUY=0, wherein the geodesic is the difference between the sum of the support and confidence of BUY=1 and the sum of the support and confidence of BUY=0, and filtering itemsets with a geodesic greater than 0 as target itemsets.

2. The method for extracting key factors of click-through rate as described in claim 1, characterized in that, Before the steps of determining a first support threshold and a first confidence threshold based on the support and confidence of each itemset in a preset click scenario, and determining a second support threshold and a second confidence threshold based on the support and confidence of each itemset in a preset non-click scenario, the method further includes: Obtain preset raw data, and perform correlation statistics based on the raw data to obtain the support, confidence and lift of each item set in the preset click scenario, as well as the support, confidence and lift of each item set in the preset non-click scenario.

3. The method for extracting key factors of click-through rate as described in claim 1, characterized in that, The step of determining the first support threshold and the first confidence threshold based on the support and confidence of each itemset under a preset click scenario includes: The support of each itemset in the click scenario is sorted to obtain the corresponding first median and first upper quantile, and the support corresponding to the first median or the support corresponding to the first upper quantile is determined as the first support threshold. The confidence scores of each itemset in the click scenario are sorted to obtain the corresponding second median and second upper quantile, and the confidence score corresponding to the second median or the confidence score corresponding to the second upper quantile is determined as the first confidence threshold.

4. The method for extracting key factors of click-through rate as described in claim 1, characterized in that, The step of determining the second support threshold and the second confidence threshold based on the support and confidence of each itemset in a preset non-click scenario includes: The support of each itemset in the non-click scenario is sorted to obtain the corresponding third median and third upper quantile, and the support corresponding to the third median or the support corresponding to the third upper quantile is determined as the second support threshold. The confidence scores of each itemset in the non-click scenario are sorted to obtain the corresponding fourth median and fourth upper quantile, and the confidence score corresponding to the fourth median or the fourth upper quantile is determined as the second confidence threshold.

5. The method for extracting key factors of click-through rate as described in claim 1, characterized in that, The steps of filtering multiple itemsets in the click scenario to obtain a first item set to be filtered based on the first support threshold, the first confidence threshold, and a preset lift threshold, and filtering multiple itemsets in the non-click scenario to obtain a second item set to be filtered based on the second support threshold and the second confidence threshold, include: From multiple item sets in the click scenario, select the item set with support greater than the first support threshold, confidence greater than the first confidence threshold, and lift greater than the preset lift threshold as the first item set to be filtered; From the multiple item sets in the non-click scenario, select the item sets with support less than the second support threshold and confidence less than the second confidence threshold as the second item set to be filtered.

6. The method for extracting key factors of click-through rate as described in claim 1, characterized in that, The step of obtaining a candidate set to be filtered based on the first and second itemsets to be filtered, and then performing geodesic filtering on the candidate sets to be filtered to obtain the target itemset, includes: The intersection of the first and second itemsets to be filtered is determined as the candidate itemset to be filtered, and geodesic filtering is performed on the candidate itemset to be filtered according to the preset itemset filtering rules to obtain the corresponding target itemset.

7. The method for extracting key factors of click-through rate as described in claim 1, characterized in that, The key factors of click-through rate (CTR) include: single CTR factor, two CTR factors, and three CTR factors. The step of extracting key factors of click-through rate from the target item set includes: Extract at least the important single factor, the important dual factor, and the important triple factor of click-through rate from the target item set.

8. A system for extracting key factors of click-through rate, characterized in that, The system for extracting key factors of click-through rate includes: The determination module is used to determine a first support threshold and a first confidence threshold based on the support and confidence of each itemset in a preset click scenario, and to determine a second support threshold and a second confidence threshold based on the support and confidence of each itemset in a preset non-click scenario. The filtering module is used to filter multiple itemsets in the click scenario to obtain a first item set to be filtered based on the first support threshold, the first confidence threshold, and a preset lift threshold, and to filter multiple itemsets in the non-click scenario to obtain a second item set to be filtered based on the second support threshold and the second confidence threshold. An extraction module is used to obtain a candidate set to be filtered based on the first and second itemsets to be filtered, perform geodesic filtering on the candidate sets to be filtered to obtain a target itemset, and extract click-through rate (CTR) important factors from the target itemset; wherein, the geodesic filtering method includes: determining whether the support and confidence corresponding to BUY=1 are higher than the support and confidence corresponding to BUY=0, wherein the geodesic is the difference between the sum of the support and confidence corresponding to BUY=1 and the sum of the support and confidence corresponding to BUY=0, and filtering itemsets corresponding to geodesics greater than 0 as target itemsets.

9. A terminal device, characterized in that, The terminal device includes a memory, a processor, and a click-through rate (CTR) important factor extraction program stored in the memory and executable on the processor. When the CTR important factor extraction program is executed by the processor, it implements the steps of the CTR important factor extraction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program for extracting click-through rate (CTR) factors, which, when executed by a processor, implements the steps of the CTR factor extraction method as described in any one of claims 1 to 7.

Images