Method and apparatus for mining user interest

Abstract

The application discloses a user interest mining method and device, and belongs to the technical field of data mining. The method comprises the following steps: acquiring a first exposure frequency of a target label of a target user; determining a first weighting coefficient of the target label and a second weighting coefficient of the target label based on the first exposure frequency, wherein the first weighting coefficient is a weighting coefficient for a target sliding window, and the second weighting coefficient is a weighting coefficient for historical storage; determining a positive feedback behavior index of the target label and an exposure index of the target label based on the first weighting coefficient and the second weighting coefficient; and determining an interest index of the target label for the target user based on the positive feedback behavior index of the target label and the exposure index of the target label; wherein the first weighting coefficient increases with the increase of a target statistical quantity in the target sliding window.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of data mining, and particularly relates to a user interest mining method and device. BACKGROUND

[0002] With the development of Internet technology, various platforms tend to automatically push information to users. However, the time and attention of users are limited, and pushing information to users without knowing their interests may cause users to be dissatisfied, damage user experience, and lead to user loss. Therefore, it is necessary to mine user interests before pushing information to users.

[0003] The prior art usually counts the number of recommendations of target information to a target user and the number of clicks of the target information by the target user, takes the ratio of the number of clicks to the number of recommendations as a historical click rate, and then mines whether the target user is interested in the target information according to the historical click rate.

[0004] However, this way of mining user interests based on historical click rates has the problem of low accuracy. SUMMARY

[0005] The embodiments of the present application provide a user interest mining method and device, which can solve the problem of low accuracy in mining user interests in related technologies.

[0006] In a first aspect, the embodiments of the present application provide a user interest mining method, which comprises:

[0007] obtaining a first exposure number of a target label of a target user, the first exposure number being an exposure number of the target label in a target sliding window;

[0008] determining a first weighting coefficient of the target label and a second weighting coefficient of the target label based on the first exposure number, the first weighting coefficient being a weighting coefficient for the target sliding window, and the second weighting coefficient being a weighting coefficient for historical storage;

[0009] determining a positive feedback behavior index of the target label and an exposure index of the target label based on the first weighting coefficient and the second weighting coefficient;

[0010] determining an interest index of the target label of the target user based on the positive feedback behavior index of the target label and the exposure index of the target label;

[0011] wherein the first weighting coefficient increases with an increase in a target statistic in the target sliding window and decreases with a decrease in the target statistic in the target sliding window.

[0012] In a second aspect, an embodiment of the present application provides a user interest mining device, which comprises:

[0013] an acquisition module, configured to acquire a first exposure frequency of a target label of a target user, the first exposure frequency being an exposure frequency of the target label in a target sliding window;

[0014] a determination module, configured to determine a first weighting coefficient of the target label and a second weighting coefficient of the target label based on the first exposure frequency, the first weighting coefficient being a weighting coefficient for the target sliding window, and the second weighting coefficient being a weighting coefficient for historical storage; determine a positive feedback behavior index of the target label based on the first weighting coefficient and the second weighting coefficient, determine an exposure index of the target label, and determine an interest index of the target label for the target label based on the positive feedback behavior index of the target label and the exposure index of the target label; wherein the first weighting coefficient increases with an increase of a target statistical quantity in the target sliding window and decreases with a decrease of the target statistical quantity in the target sliding window.

[0015] In a third aspect, an embodiment of the present application provides an electronic device, which comprises a processor and a memory, the memory storing a program or instruction, the program or instruction being executed by the processor to implement the steps of the method according to the first aspect.

[0016] In a fourth aspect, an embodiment of the present application provides a readable storage medium, the readable storage medium storing a program or instruction, the program or instruction being executed by a processor to implement the steps of the method according to the first aspect.

[0017] In a fifth aspect, an embodiment of the present application provides a chip, which comprises a processor and a communication interface, the communication interface being coupled with the processor, and the processor being configured to run a program or instruction to implement the method according to the first aspect.

[0018] In a sixth aspect, an embodiment of the present application provides a computer program product, which is stored in a storage medium and is executed by at least one processor to implement the method according to the first aspect.

[0019] In the embodiment of the present application, by acquiring a first exposure frequency of a target label of a target user, the first exposure frequency being an exposure frequency of the target label in a target sliding window; based on the first exposure frequency, determining a first weighting coefficient of the target label and a second weighting coefficient of the target label, the first weighting coefficient being a weighting coefficient for the target sliding window; based on the first weighting coefficient and the second weighting coefficient, determining a positive feedback behavior index of the target label and an exposure index of the target label; based on the positive feedback behavior index of the target label and the exposure index of the target label, determining an interest index of the target user for the target label; wherein the first weighting coefficient increases with an increase of a target statistic quantity in the target sliding window and decreases with a decrease of the target statistic quantity in the target sliding window. In this way, by introducing the first weighting coefficient which increases with an increase of the target statistic quantity in the target sliding window and the second weighting coefficient which decreases with a decrease of the target statistic quantity in the target sliding window, not only can the stored data in the sliding window and the historical stored data be considered when mining user interest, but also the influence weight of the stored data in the target sliding window and the historical stored data on the interest index can be dynamically adjusted, avoiding that the interest index always reflects the historical interest of the user, and the accuracy of user interest mining can be improved to a certain extent. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a flowchart of a user interest mining method provided by an embodiment of the present application;

[0021] Figure 2 is a flowchart of another user interest mining method provided by an embodiment of the present application;

[0022] Figure 3 is a flowchart of another user interest mining method provided by an embodiment of the present application;

[0023] Figure 4 is a structural block diagram of a user interest mining device provided by an embodiment of the present application;

[0024] Figure 5 is a structural block diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be clearly described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art belong to the scope of protection of the present application.

[0026] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally a class, and do not limit the number of objects, for example, the first object can be one or more. In addition, "and / or" in the specification and claims indicates at least one of the connected objects, and the character " / ", generally indicates that the front and rear associated objects are in a "or" relationship.

[0027] The user interest mining method provided by the embodiments of the present application will be described in detail below in combination with the drawings, specific embodiments and application scenarios.

[0028] Figure 1 is a flowchart of a user interest mining method provided by an embodiment of the present application. The user interest mining method provided by the embodiments of the present application can be executed by an electronic device, which can be, for example, a mobile phone, a personal computer, a server, a smart wearable device, etc. As shown in Figure 1 The user interest mining method provided by the embodiments of the present application can include the following steps:

[0029] Step 110, obtaining a first exposure frequency of a target label of a target user, the first exposure frequency being an exposure frequency of the target label in a target sliding window;

[0030] The target label can be a feature of an object (for example, a commodity or a video) acted on by the target user behavior, and the target label can be, for example, a fitness label, a food label, a travel label, etc. For example, each video on a video website can have some labels, for example, a jumping exercise video can have a weight loss label and a fitness label, a snack review video can have a food label and a sweet food label, and similarly, each commodity on a shopping platform can also have various labels. The label possessed by the video or commodity acted on by the target user behavior is the target label. The exposure frequency can reflect the behavior frequency of the target user to the object carrying the target label.

[0031] The target sliding window can be a fixed-length window, and the specific length can be set according to actual needs. The target sliding window can be used to store the behaviors of the target user in the recent period (for example, the last week, the last three days, or the last month, etc.), and the statistical quantity related to the target label in the target user behaviors in the recent period is counted. The format of the data stored in the target sliding window can be, for example, target user ID, label number, and one user behavior. Meanwhile, the one user behavior can include at least one of a browsing behavior or a positive feedback behavior. The browsing behavior can be divided into a behavior with a browsing time exceeding a target time (for example, 3 seconds) and an implicit negative feedback behavior with a browsing time not exceeding a target time (for example, 3 seconds), and the positive feedback behavior can include a click, a forward, a comment, a like, and the like. It should be noted that in the embodiments of the present application, the implicit negative feedback is a relative concept of the explicit negative feedback. The explicit negative feedback, which is often referred to as negative feedback, refers to a user actively using the complaint, shielding, and the like functions of a website or an application program to explicitly feedback the content that he or she dislikes. The implicit negative feedback refers to a user not actively using the complaint, shielding, and the like functions of a website or an application program, but indirectly feedbacking the content that he or she dislikes through indirect behaviors (for example, shortening the browsing time, etc.).

[0032] The statistical quantity can include the exposure times of the target label in the target sliding window, the positive feedback behavior times of the target label in the target sliding window, and the implicit negative feedback index of the target label in the target sliding window. The exposure times of the target label in the target sliding window can represent the total number of target user behaviors corresponding to the target label stored in the target sliding window. The positive feedback behavior times of the target label in the target sliding window can represent the total number of target user behaviors corresponding to the target label and the positive feedback behavior stored in the target sliding window. The implicit negative feedback index of the target label in the target sliding window can represent data determined according to the positive feedback behavior times and the implicit negative feedback behavior times in the target sliding window. In the embodiments of the present application, the initial values of the statistical quantities in the target sliding window can all be 0.

[0033] Taking a target sliding window with a fixed length of 200 as an example, the target sliding window can store the last 200 behaviors of the target user. The target user behaviors exceeding the storage length of the target sliding window can be retired from the target sliding window according to the first-in first-out principle and transferred to the historical storage. In the target sliding window, the total number of times of the target label appearing is the first exposure times of the target label.

[0034] In step 120, a first weighting coefficient of the target label and a second weighting coefficient of the target label are determined based on the first exposure frequency, the first weighting coefficient being a weighting coefficient for the target sliding window, and the second weighting coefficient being a weighting coefficient for historical storage; wherein the first weighting coefficient increases with an increase in a target statistic in the target sliding window and decreases with a decrease in the target statistic in the target sliding window.

[0035] The first weighting coefficient and the second weighting coefficient can be real numbers greater than or equal to 0 and less than or equal to 1.

[0036] The first weighting coefficient can be a real number obtained by mapping the first exposure frequency to the interval [0, 1], or a real number obtained by mapping the first exposure frequency to the interval [0, 1] and further multiplying a preset coefficient (for example, 0.5, 0.6). Based on the first exposure frequency, the first weighting coefficient of the target label can be determined, which can include: inputting the first exposure frequency into an activation function to obtain the first weighting coefficient of the target label, or inputting the first exposure frequency into an activation function to obtain a first value, and obtaining the first weighting coefficient of the target label according to the product of the first value and a preset coefficient; wherein the preset coefficient is greater than zero and less than one. In the embodiments of the present application, the activation function can include at least one of a Sigmoid function, a tanh function, and a Relu function. For example, the first weighting coefficient can be equal to Sigmoid (first exposure frequency) or k*Sigmoid (first exposure frequency), where k is a preset coefficient.

[0037] The target statistic in the target sliding window can include the first exposure frequency or the number of positive feedback behaviors in the target sliding window. Taking the target statistic as the first exposure frequency as an example, the first weighting coefficient can increase with the increase of the first exposure frequency and decrease with the decrease of the first exposure frequency. At the same time, the second weighting coefficient can decrease with the increase of the first exposure frequency and increase with the decrease of the first exposure frequency.

[0038] In step 130, a positive feedback behavior indicator of the target label is determined based on the first weighting coefficient and the second weighting coefficient, and an exposure indicator of the target label is determined.

[0039] In the embodiments of the present application, the positive feedback behavior indicator can be an indicator for reflecting the overall situation of the number of positive feedback behaviors of the target label in the target sliding window and the historical storage. The exposure indicator can be an indicator for reflecting the overall situation of the number of exposures of the target label in the target sliding window and the historical storage. The specific form of the positive feedback behavior indicator and the exposure indicator can be a score. Based on the first weighting coefficient and the second weighting coefficient, determining the positive feedback behavior indicator of the target label can include: determining the number of positive feedback behaviors of the target label in the target sliding window, determining the number of positive feedback behaviors of the target label in the historical storage; combining the product of the first weighting coefficient and the number of positive feedback behaviors in the target sliding window and the product of the second weighting coefficient and the number of positive feedback behaviors in the historical storage, to determine the positive feedback behavior indicator of the target label, or amplifying the number of positive feedback behaviors of the target label in the target sliding window (for example, multiplying the number of positive feedback behaviors in the target sliding window by a real number greater than or equal to 2), and combining the product of the first weighting coefficient and the amplified number of positive feedback behaviors in the target sliding window and the product of the second weighting coefficient and the number of positive feedback behaviors in the historical storage, to determine the positive feedback behavior indicator of the target label. The exposure indicator of the target label can be the number of exposures of the target label in the historical storage or the sum of the number of exposures of the target label in the historical storage and the first number of exposures.

[0040] Step 140, based on the positive feedback behavior indicator of the target label and the exposure indicator of the target label, determining the interest indicator of the target user for the target label.

[0041] The interest indicator can be a score determined according to the overall situation of the target label in the target sliding window and the historical storage, for reflecting the degree of interest of the target user in the target label. Based on the positive feedback behavior indicator of the target label and the exposure indicator of the target label, determining the interest indicator of the target user for the target label can include: inputting the positive feedback behavior indicator and the exposure indicator into the Wilson formula to determine the interest indicator of the target user for the target label.

[0042] A threshold value can be set for the interest indicator. When the interest indicator is higher than the threshold value, it can be determined that the target user is more interested in the target label. When the interest indicator is not higher than the threshold value, it can be determined that the target user is less interested in the target label. In the case where the interest indicator of the target user for the target label is higher than the threshold value, more related content with the target label can be pushed to the user. In the case where the interest indicator of the target user for the target label is not higher than the threshold value, the pushing of related content with the target label can be reduced.

[0043] The user interest mining method provided by the embodiments of the present application can not only consider the stored data in the sliding window and the historical stored data when mining the user interest, but also dynamically adjust the influence weight of the stored data in the target sliding window and the historical stored data on the interest index, thereby avoiding the interest index from always reflecting the historical interest of the user and improving the accuracy of the user interest mining to a certain extent.

[0044] In an embodiment of the present application, the target sliding window is used to store a preset number of user behaviors for the target user in chronological order, and user behaviors exceeding the preset number are located in the historical storage; the preset number of user behaviors in the target sliding window follows the first-in first-out principle; if a user behavior for the target label is newly added in the target sliding window, the first exposure number is increased by one statistical unit; if a user behavior for the target label is reduced in the target sliding window, the first exposure number is reduced by one statistical unit; if a positive feedback behavior for the target label is newly added in the target sliding window, the positive feedback behavior number of the target label is increased by one statistical unit; if a positive feedback behavior for the target label is reduced in the target sliding window, the positive feedback behavior number of the target label is reduced by one statistical unit.

[0045] In the embodiments of the present application, the preset number can be set according to requirements, and can be, for example, 200, 1000, etc. The target sliding window can store a preset number of user behaviors for the target user in chronological order at most. With the passage of time and the increase of the target user behaviors, user behaviors exceeding the preset number can be retired from the target sliding window into the historical storage according to the first-in first-out principle. The positive feedback behavior number of the target label can be the positive feedback number of the target label in the target sliding window. One statistical unit can be, for example, 1, 5, 6, etc.

[0046] When the target user behavior enters the target sliding window, if it is detected that a user behavior for the target label is newly added in the target sliding window (for example, it is detected that the label number corresponding to the newly added user behavior in the target sliding window is the same as the target label number), the first exposure number can be increased by one statistical unit. With the passage of time, when the target user behavior retires from the target sliding window, if it is detected that a user behavior for the target label is reduced in the target sliding window (for example, it is detected that the label number corresponding to the user behavior retiring from the target sliding window is the same as the target label number), the first exposure number can be reduced by one statistical unit.

[0047] When the target user behavior enters the target sliding window, if an increase in the target user behavior for the target label in the target sliding window is detected, and the target user behavior for the target label is a positive feedback behavior, the number of positive feedback behaviors of the target label can be increased by one statistical unit. When the target user behavior exits the target sliding window, if a decrease in the target user behavior for the target label in the target sliding window is detected, and the target user behavior for the target label is a positive feedback behavior, the number of positive feedback behaviors of the target label can be decreased by one statistical unit.

[0048] In this way, the target user behaviors are stored in the target sliding window, and the first exposure number and the number of positive feedback behaviors of the target label are adjusted as the target user behaviors change, so that the calculated interest index changes as the data stored in the target sliding window changes, thereby more sensitively reflecting changes in the interest of the target user.

[0049] In an embodiment of the present application, after the interest index of the target user for the target label is determined based on the positive feedback behavior index of the target label and the exposure index of the target label in step 140, the user interest mining method can further include: in the case where the interest index of the target label is greater than a target threshold, adding the target label to a set of interest labels of the target user; and pushing content corresponding to each label in the set of interest labels to the target user. In this way, the content corresponding to the interest labels can be pushed to the target user in a targeted manner, achieving accurate pushing.

[0050] The set of interest labels of the target user can be a set of labels that the target user is interested in. The target threshold can be a standard for measuring whether the target user is interested in the target label. When the interest index of the target label is greater than the target threshold, it can be considered that the target user is interested in the target label, and the target label can be added to the set of interest labels of the target user at this time. After the target label is added to the set of interest labels of the target user, only content corresponding to each label in the set of interest labels can be pushed to the target user, avoiding pushing content that the target user is not interested in, and causing user dissatisfaction.

[0051] When the interest index of the target label is not higher than the target threshold, it can be considered that the target user is not interested in the target label, and the target label can be ignored and no related content is pushed.

[0052] Of course, it needs to be explained that when the target label is added to the target user's interest label set, the interest index of the target label can be further recorded. Thus, when the target user is pushed with the content corresponding to each label in the interest label set, the interest index can be combined for pushing. Specifically, the interest index of each label in the interest label set can be used to sort the labels, and for the labels with high interest index (i.e., high ranking), the pushing frequency or the pushing times of the content corresponding to the labels can be increased.

[0053] Figure 2 is a flowchart of a user interest mining method provided by an embodiment of the present application. As shown in Figure 2 the user interest mining method provided by an embodiment of the present application can include the following steps:

[0054] In step 210, the first exposure times of the target label of the target user are obtained, the first exposure times being the exposure times of the target label in the target sliding window;

[0055] In step 220, based on the first exposure times, the first weighting coefficient of the target label and the second weighting coefficient of the target label are determined, the first weighting coefficient being the weighting coefficient for the target sliding window, and the second weighting coefficient being the weighting coefficient for the historical storage; wherein the first weighting coefficient increases with the increase of the target statistical quantity in the target sliding window, and decreases with the decrease of the target statistical quantity in the target sliding window;

[0056] In step 230, the first positive feedback behavior times and the second positive feedback behavior times of the target label of the target user are obtained; the first positive feedback behavior times being the positive feedback behavior times of the target label in the target sliding window, and the second positive feedback behavior times being the positive feedback behavior times of the target label in the historical storage;

[0057] The first positive feedback behavior times can be the total times of the positive feedback behavior of the target label by the target user in the target sliding window. The second positive feedback behavior times can be the total times of the positive feedback behavior of the target label by the target user in the historical storage. In the present embodiment, the data format of the data stored in the target sliding window and the data stored in the historical storage can be the same. As described above, the format of the data stored in the target sliding window can be, for example, target user ID, label number, and one-time user behavior. In this case, the first positive feedback behavior times can be the total number of the user behaviors containing the target label number and the one-time user behavior being the positive feedback behavior stored in the target interaction window. The second positive feedback behavior times can be the total number of the user behaviors containing the target label number and the one-time user behavior being the positive feedback behavior stored in the historical storage.

[0058] Since the user interest changes over time, the data in the history storage has less influence on the user interest, in the embodiment of the present application, the second positive feedback behavior times can also be determined based on the exposure times of the target label in the history storage and a time decay factor.

[0059] In step 240, the second exposure times of the target label of the target user are obtained, the second exposure times being the exposure times of the target label in the history storage;

[0060] The second exposure times can be the total times of the behavior of the target user for the target label stored in the history storage. Since the user interest changes over time, the data in the history storage has less influence on the user interest, in the embodiment of the present application, the second exposure times can also be determined based on the exposure times of the target label in the history storage and a time decay factor.

[0061] In step 250, an amplification coefficient of the target statistic quantity for the target sliding window is determined.

[0062] The target statistic quantity can include at least one of the first exposure times or the first positive feedback times. The amplification coefficient can be used to amplify the target statistic quantity, and flatten the order of magnitude of the target statistic quantity in the target sliding window and the corresponding data in the history storage. In the embodiment of the present application, the amplification coefficient can be determined based on the first exposure times and the second exposure times, for example, the ratio of the first exposure times to the second exposure times is taken as the amplification coefficient, or the sum of the ratio of the first exposure times to the second exposure times and a constant function is taken as the amplification coefficient, or a preset value is directly taken as the amplification coefficient.

[0063] In step 260, based on the first weighting coefficient, the first positive feedback behavior times, the amplification coefficient, the second weighting coefficient and the second positive feedback behavior times, a positive feedback behavior index of the target label of the target user is determined.

[0064] In the embodiment of the present application, the positive feedback behavior index can have multiple expression forms. The first kind is positive feedback behavior index = first weighting coefficient * (first positive feedback times / amplification coefficient) + second weighting coefficient * second positive feedback behavior times. The second kind is positive feedback behavior index = first weighting coefficient * (first positive feedback times / amplification coefficient) * constant function + second weighting coefficient * second positive feedback behavior times.

[0065] In step 270, based on the first weighting coefficient, the first exposure times, the amplification coefficient, the second weighting coefficient and the second exposure times, an exposure index of the target label of the target user is determined, or the second exposure times are taken as the exposure index of the target label of the target user.

[0066] In the embodiments of the present application, the exposure index can have various forms of expression. The first form is exposure index = first weighting coefficient * (first exposure frequency / magnification factor) + second weighting coefficient * second exposure frequency. The second form is exposure index = first weighting coefficient * (first exposure frequency / magnification factor) * constant function + second weighting coefficient * second exposure frequency. The third form is exposure index = second exposure frequency.

[0067] In step 280, the interest index of the target user for the target label is determined based on the positive feedback behavior index of the target label and the exposure index of the target label.

[0068] The method for mining user interest provided in the embodiments of the present application can magnify the target statistical quantity in the target sliding window by introducing a magnification factor, which can avoid the situation that the quantity order of the target statistical quantity in the target sliding window is much smaller than the quantity order of the related data in the historical storage, so that the target statistical quantity in the target sliding window is always at a disadvantage when the interest index is calculated by merging the target statistical quantity in the target sliding window and the historical data. On the other hand, the magnification factor can magnify the effect of the target sliding window, so that the effect of the target sliding window and the historical storage can be leveled when the target statistical quantity in the target sliding window is large and the confidence is high, thereby improving the accuracy of the interest mining result.

[0069] In one embodiment of the present application, step 260 can determine the positive feedback behavior index of the target label of the target user based on the first weighting coefficient, the first positive feedback behavior frequency, the magnification factor, the second weighting coefficient and the second positive feedback behavior frequency, which can include B = W1 * (N1 / R) + W2 * N2, where B is the positive feedback behavior index, W1 is the first weighting coefficient, N1 is the first positive feedback behavior frequency, R is the magnification factor, W2 is the second weighting coefficient, and N2 is the second positive feedback behavior frequency. Step 270 can determine the exposure index of the target label of the target user based on the first weighting coefficient, the first exposure frequency, the magnification factor, the second weighting coefficient and the second exposure frequency, which can include calculating the exposure index of the target label of the target user by E = W1 * (M1 / R) + W2 * M2, where E is the exposure index, W1 is the first weighting coefficient, M1 is the first exposure frequency, R is the magnification factor, W2 is the second weighting coefficient, and M2 is the second exposure frequency; and the sum of the second weighting coefficient and the first weighting coefficient is 1.

[0070] In an embodiment of the present application, the determining the amplification coefficient of the target statistical quantity of the target sliding window comprises: taking a ratio of the first exposure times and the second exposure times as the amplification coefficient of the target statistical quantity of the target sliding window; or taking a preset value as the amplification coefficient of the target statistical quantity of the target sliding window.

[0071] For the first exposure times, the second exposure times, the first positive feedback behavior times and the second positive feedback behavior times mentioned above, the present application further introduces the determining process thereof. In an embodiment of the present application, the target sliding window is used to store a preset number of user behaviors for the target user in chronological order, and user behaviors exceeding the preset number are located in the historical storage; the user behaviors in the target sliding window follow the first-in first-out principle. If a user behavior for the target label is newly added in the target sliding window, the first exposure times is increased by one statistical unit; if a user behavior for the target label is reduced in the target sliding window, the first exposure times is reduced by one statistical unit, and the second exposure times is attenuated over time and increased by one statistical unit. If a positive feedback behavior for the target label is newly added in the target sliding window, the first positive feedback behavior times of the target label is increased by one statistical unit; if a positive feedback behavior for the target label is reduced in the target sliding window, the first positive feedback behavior times of the target label is reduced by one statistical unit, and the second positive feedback behavior times is attenuated over time and increased by one statistical unit.

[0072] Taking one statistical unit as 1 for example, if a user behavior for the target label is newly added in the target sliding window, the first exposure times can be updated, and the updated first exposure times can be equal to the first exposure times before the update plus 1; if the user behavior for the target label is a positive feedback behavior, the first positive feedback behavior times can be updated, and the updated first positive feedback behavior times can be equal to the first positive feedback behavior times before the update plus 1.

[0073] If the user behavior for the target label in the target sliding window is reduced, the first exposure times can be updated, and the updated first exposure times can be equal to the first exposure times before the update minus 1. If the user behavior for the target label is a positive feedback behavior, the first positive feedback behavior times can be updated, and the updated first positive feedback behavior times can be equal to the first positive feedback behavior times before the update minus 1. At the same time, since the user behavior for the target label in the target sliding window is reduced and enters the historical storage, the second exposure times can be updated, and the updated second exposure times can be equal to the second exposure times before the update multiplied by the time decay factor and added to 1 (i.e., the updated second exposure times = the second exposure times before the update * the time decay factor + 1). If the user behavior for the target label is a positive feedback behavior, the second positive feedback behavior times can be updated, and the updated second positive feedback behavior times can be equal to the second positive feedback behavior times before the update multiplied by the time decay factor and added to 1 (i.e., the updated second positive feedback behavior times = the second positive feedback behavior times before the update * the time decay factor + 1).

[0074] Since the user interest changes over time, the user's recent behavior usually reflects the user's interest, and the related art provides a way of calculating a score based on the current score and the history score (determined by the product of the historical storage statistics and the time decay factor) to mine the user interest. However, this way has the following problems: 1. When the time decay factor is too large, the user's historical information is lost, resulting in that the score is basically completely dependent on the current behavior, and the number of current behaviors is much smaller than the number of historical behaviors, and the small number of behaviors is difficult to guarantee the accuracy in a statistical sense. 2. When the time decay factor is too small, the historical behavior has a large data volume and a high score proportion, and the user's current behavior is difficult to affect the final score, resulting in that the system lacks sensitivity and cannot update the interest score in time according to the user's recent behavior feedback.

[0075] The user interest mining method provided by the embodiments of the present application can store the recent behavior of a target user by using a target sliding window, and dynamically adjust the influence weight of the stored data in the target sliding window and the historical stored data on an interest index by using a first weighting coefficient that increases with the increase of a target statistic quantity in the target sliding window and a second weighting coefficient that is 1 minus the first weighting coefficient. In this way, on the one hand, when the target statistic quantity in the target sliding window is insufficient and the confidence is not high, the influence weight of the historical stored data on the interest index is amplified by the second weighting coefficient, and the accuracy of the interest mining result is ensured based on a large amount of historical stored data; on the other hand, when the target statistic quantity in the target sliding window increases, the influence weight of the stored data in the target sliding window on the interest index is amplified by the first weighting coefficient, and the recent behavior of the user is timely reflected in the interest index. Compared with the method of determining the user interest based on only a time decay factor, the method is not only more accurate but also more sensitive.

[0076] Figure 3 is a flowchart of another user interest mining method provided by the embodiments of the present application. As shown in Figure 3 the user interest mining method provided by the embodiments of the present application can include the following steps:

[0077] Step 310, obtaining a first exposure frequency of a target label of a target user, the first exposure frequency being an exposure frequency of the target label in a target sliding window;

[0078] Step 320, determining a first weighting coefficient of the target label and a second weighting coefficient of the target label based on the first exposure frequency, the first weighting coefficient being a weighting coefficient for the target sliding window, and the second weighting coefficient being a weighting coefficient for historical storage; wherein the first weighting coefficient increases with the increase of a target statistic quantity in the target sliding window, and decreases with the decrease of the target statistic quantity in the target sliding window;

[0079] Step 330, determining a positive feedback behavior index of the target label and an exposure index of the target label based on the first weighting coefficient and the second weighting coefficient;

[0080] Step 340, determining an interest index of the target label of the target user based on the positive feedback behavior index of the target label and the exposure index of the target label;

[0081] Step 350, screening the target label from an interest label set of the target user in a case where the target label meets a target condition;

[0082] In the embodiments of the present application, the interest label set of the target user can be a set of labels of interest to the target user. The labels of interest to the target user can be pre-set default labels, and / or a part of labels selected according to interest indicators of the labels (for example, target labels with interest indicators higher than a target threshold). The target condition can include at least two of the following: the value of the first exposure frequency is greater than or equal to a first threshold; the implicit negative feedback index of the target label of the target user in the target sliding window is greater than a second threshold; and the click rate of the target label of the target user in the target sliding window is lower than the average click rate of the target user in the target sliding window.

[0083] The first threshold can be, for example, a positive integer greater than or equal to 3. The second threshold can be, for example, a positive integer greater than or equal to 0. The implicit negative feedback index can be data determined according to the number of positive feedback behaviors and the number of implicit negative feedback behaviors of the target label in the target sliding window, wherein the implicit negative feedback index can decrease with an increase in the number of positive feedback behaviors of the target label in the target sliding window, and increase with an increase in the number of implicit negative feedback behaviors of the target label in the target sliding window. The click rate of the target label of the target user in the target sliding window can be equal to the ratio of the number of positive feedback behaviors of the target label in the target sliding window (the first number of positive feedback behaviors) to the first exposure frequency. The average click rate of the target user in the target sliding window can be the ratio of the number of positive feedback behaviors of all labels in the target sliding window to the total number of user behaviors stored in the target sliding window.

[0084] Step 360, in the case that the target label is in a shielding state, the content corresponding to the target label is not pushed to the target user.

[0085] The user interest mining method provided in the embodiments of the present application can shield the target label from the interest labels of the target user in the case that the target label meets the target condition, which can shield the target label in a short period of time, and does not affect the stored data in the target sliding window and the statistics related to the target label, thereby avoiding irreversible modification of the stored data in the target sliding window and the statistics related to the target label, which can permanently damage the accuracy of user interest mining, and ensuring the accuracy of the stored data in the target sliding window and the statistics related to the target label.

[0086] In the embodiments of the present application, whether the target label needs to be shielded from the interest label set of the target user can be determined from multiple dimensions. In one embodiment of the present application, the target condition includes at least two of the following: the value of the first exposure frequency is greater than or equal to a set number threshold; the implicit negative feedback index of the target label of the target user in the target sliding window is greater than 0; wherein the implicit negative feedback index is an index obtained based on the indirect behavior of the target user, and is used to reflect the degree of disinterest of the target user in the target label; the click rate of the target user on the target label in the target sliding window is lower than the average click rate of the target user in the target sliding window. In this way, whether the target label meets the target condition can be determined by using the implicit negative feedback index, and the target label is directly shielded from the user interest label set when the target condition is met. On the one hand, the implicit negative feedback index is reasonably utilized, and on the other hand, compared with the way of bringing the implicit negative feedback index into the calculation of the interest index, the noise in the implicit negative feedback behavior of the user can be avoided to be brought into the user interest index, and thus the accuracy of user interest mining can be improved to a certain extent.

[0087] The set number threshold can be a positive integer greater than or equal to 3, and the specific value can be determined according to actual conditions. The indirect behavior can be the implicit negative feedback behavior mentioned above, for example, a behavior of less than 3s of browsing time. In the embodiments of the present application, the implicit negative feedback index can be obtained only according to the number of implicit negative feedback behaviors of the target label in the target sliding window. The implicit negative feedback index can also be obtained in combination with the number of first positive feedback behaviors and the number of implicit negative feedback behaviors of the target label in the target sliding window.

[0088] It can be understood that the more the number of implicit negative feedback behaviors of the target label in the target sliding window, the higher the implicit negative feedback index, and the more the target user is disinterested in the target label. With the update of data in the target sliding window, if the number of first positive feedback behaviors increases, the interest of the target user can change, and thus the implicit negative feedback index can be appropriately reduced according to the number of first positive feedback behaviors, the implicit negative feedback index is corrected, so that the implicit negative feedback index can better reflect the degree of disinterest of the target user in the target label.

[0089] In one embodiment of the present application, after step 350, the user interest mining method provided in the embodiments of the present application can further include: in the case where the target label does not meet the target condition, the target label is unshielded from the interest label set of the target user.

[0090] The case that the target label does not satisfy the target condition can include at least one of the following: the number of the first exposure times is less than a set number threshold; the implicit negative feedback index of the target label of the target user in the target sliding window is not greater than 0; and a click rate of the target label of the target user in the target sliding window is not less than an average click rate of the target user in the target sliding window. After the target label is unshielded from the interest label set of the target user, content corresponding to the target label can continue to be pushed to the target user.

[0091] In an embodiment of the present application, in the case that the target condition includes that the implicit negative feedback index of the target label of the target user in the target sliding window is greater than 0, the implicit negative feedback index increases with an increase in user implicit negative feedback behavior for the target label in the target sliding window and decreases with an increase in user positive feedback behavior for the target label in the target sliding window; and the number of statistical units for which the implicit negative feedback index decreases for one user positive feedback behavior is greater than the number of statistical units for which the implicit negative feedback index increases for one user implicit negative feedback behavior. In this way, the implicit negative feedback index can be corrected by user positive feedback behavior for the target label in the target sliding window, so that the implicit negative feedback index changes rapidly with changes in user interest, and the accuracy of interest mining can be improved relative to a manner of determining the implicit negative feedback index only according to negative implicit feedback behavior.

[0092] The initial value of the implicit negative feedback index can be 0. The implicit negative feedback index can be corrected by user positive feedback behavior for the target label. That is, the implicit negative feedback index can be updated correspondingly each time user positive feedback behavior for the target label increases in the target sliding window, and decreases with an increase in user positive feedback behavior for the target label. Taking an example in which the number of statistical units for which the implicit negative feedback index decreases for one user positive feedback behavior is 5 and the number of statistical units for which the implicit negative feedback index increases for one user implicit negative feedback behavior is 1, each time user positive feedback behavior for the target label increases in the target sliding window, the implicit negative feedback index can be updated, and the updated implicit negative feedback index can be equal to the implicit negative feedback index before the update minus 5. Each time user implicit negative feedback behavior for the target label increases in the target sliding window, the implicit negative feedback index can be updated, and the updated implicit negative feedback index can be equal to the implicit negative feedback index before the update plus 1. In addition, each time user positive feedback behavior for the target label decreases in the target sliding window, the implicit negative feedback index can increase by 5, and each time user implicit negative feedback behavior for the target label decreases in the target sliding window, the implicit negative feedback index can decrease by 1. It should be noted that the number of statistical units described above is only used for explanation and illustration, and cannot be regarded as a limitation on the present application.

[0093] Figure 4is a structural block diagram of a user interest mining device provided by an embodiment of the present application. As shown in Figure 4 The user interest mining device 400 provided by the embodiment of the present application includes an acquisition module 410, a determination module 420,

[0094] The acquisition module 410 is configured to acquire a first exposure frequency of a target label of a target user, the first exposure frequency being an exposure frequency of the target label in a target sliding window.

[0095] The determination module 420 is configured to determine a first weighting coefficient of the target label and a second weighting coefficient of the target label based on the first exposure frequency, the first weighting coefficient being a weighting coefficient for the target sliding window, and the second weighting coefficient being a weighting coefficient for historical storage; determine a positive feedback behavior index of the target label and an exposure index of the target label based on the first weighting coefficient and the second weighting coefficient; and determine an interest index of the target label of the target user based on the positive feedback behavior index of the target label and the exposure index of the target label; wherein the first weighting coefficient increases with an increase of a target statistical quantity in the target sliding window and decreases with a decrease of the target statistical quantity in the target sliding window.

[0096] The user interest mining device provided by the embodiment of the present application introduces the first weighting coefficient which increases with an increase of a target statistical quantity in a target sliding window and the second weighting coefficient which decreases with a decrease of the target statistical quantity in the target sliding window, so that not only the stored data in the sliding window and the historical stored data can be considered when the user interest is mined, but also the influence weight of the stored data in the target sliding window and the historical stored data on the interest index can be dynamically adjusted, thereby avoiding that the interest index always reflects the historical interest of the user and improving the accuracy of the user interest mining to a certain extent.

[0097] Optionally, in an embodiment of the present application, the acquisition module 410 is further configured to acquire a first positive feedback behavior number and a second positive feedback behavior number of the target label of the target user; the first positive feedback behavior number is a positive feedback behavior number of the target label in a target sliding window storage, and the second positive feedback behavior number is a positive feedback behavior number of the target label in a historical storage; acquire a second exposure number of the target label of the target user, the second exposure number being an exposure number of the target label in the historical storage; the determination module 420 is further configured to determine an amplification coefficient of a target statistical quantity for the target sliding window; in the process of determining the positive feedback behavior index of the target label based on the first weighting coefficient and the second weighting coefficient, the determination module 420 is specifically configured to determine the positive feedback behavior index of the target label of the target user based on the first weighting coefficient, the first positive feedback behavior number, the amplification coefficient, the second weighting coefficient, and the second positive feedback behavior number; in the process of determining the exposure index of the target label based on the first weighting coefficient and the second weighting coefficient, the determination module 420 is specifically configured to determine the exposure index of the target label of the target user based on the first weighting coefficient, the first exposure number, the amplification coefficient, the second weighting coefficient, and the second exposure number, or take the second exposure number as the exposure index of the target label of the target user.

[0098] Optionally, in an embodiment of the present application, in the process of determining the positive feedback behavior index of the target label of the target user based on the first weighting coefficient, the first positive feedback behavior number, the amplification coefficient, the second weighting coefficient, and the second positive feedback behavior number, the determination module 420 is specifically configured to calculate the positive feedback behavior index of the target label of the target user by: B = W1*(N1 / R) + W2*N2, wherein B is the positive feedback behavior index, W1 is the first weighting coefficient, N1 is the first positive feedback behavior number, R is the amplification coefficient, W2 is the second weighting coefficient, and N2 is the second positive feedback behavior number; in the process of determining the exposure index of the target label of the target user based on the first weighting coefficient, the first exposure number, the amplification coefficient, the second weighting coefficient, and the second exposure number, the determination module 420 is specifically configured to calculate the exposure index of the target label of the target user by: E = W1*(M1 / R) + W2*M2, wherein E is the exposure index, W1 is the first weighting coefficient, M1 is the first exposure number, R is the amplification coefficient, W2 is the second weighting coefficient, and M2 is the second exposure number; the sum of the second weighting coefficient and the first weighting coefficient is 1.

[0099] Optionally, in an embodiment of the present application, in the process of determining the amplification coefficient of the statistical quantity of the target sliding window, the determining module 420 is specifically configured to: take the ratio of the first exposure times and the second exposure times as the amplification coefficient of the target statistical quantity of the target sliding window; or take a preset value as the amplification coefficient of the target statistical quantity of the target sliding window.

[0100] Optionally, in an embodiment of the present application, the target sliding window is configured to store a preset number of user behaviors for the target user in chronological order, and user behaviors exceeding the preset number are located in the historical storage; the preset number of user behaviors in the target sliding window follow the first-in first-out principle; if a user behavior for the target label is newly added in the target sliding window, the first exposure times increases by one statistical unit; if a user behavior for the target label is reduced in the target sliding window, the first exposure times decreases by one statistical unit; if a positive feedback behavior for the target label is newly added in the target sliding window, the positive feedback behavior index of the target label increases by one statistical unit; if a positive feedback behavior for the target label is reduced in the target sliding window, the positive feedback behavior index of the target label decreases by one statistical unit.

[0101] Optionally, in an embodiment of the present application, the user interest mining apparatus 400 further comprises a processing module, which is configured to: in the case that the target label meets a target condition, shield the target label from the interest label set of the target user; and in the case that the target label is in a shielded state, not push content corresponding to the target label to the target user.

[0102] Optionally, in an embodiment of the present application, the target condition comprises at least two of the following: the first exposure times is greater than or equal to a set number threshold; the implicit negative feedback index of the target label of the target user in the target sliding window storage is greater than 0; wherein the implicit negative feedback index is an index obtained based on the indirect behavior of the target user, and is used to reflect the degree of disinterest of the target user in the target label; the click rate of the target label of the target user in the target sliding window is lower than the average click rate of the target user in the target sliding window.

[0103] Optionally, in an embodiment of the present application, the processing module is further configured to: in the case that the target label does not meet the target condition, unshield the target label from the interest label set of the target user.

[0104] Optionally, in an embodiment of the present application, when the target condition includes that the implicit negative feedback index of the target tag of the target user in the target sliding window storage is greater than 0, the implicit negative feedback index increases with the increase of the implicit negative feedback behavior of the user in the target sliding window, and decreases with the increase of the positive feedback behavior of the user in the target sliding window; the implicit negative feedback index is greater than the number of statistical units for once user positive feedback behavior decrease, and is less than the number of statistical units for once user implicit negative feedback behavior increase.

[0105] Optionally, in an embodiment of the present application, after determining the interest index of the target user to the target tag based on the positive feedback behavior index of the target tag and the exposure index of the target tag, the processing module is further configured to: add the target tag to the interest tag set of the target user when the interest index of the target tag is greater than a target threshold; and push content corresponding to each tag in the interest tag set to the target user.

[0106] It should be noted that the user interest mining device provided by the embodiments of the present application corresponds to the user interest mining method mentioned above. For related content, refer to the description of the user interest mining method above, which will not be repeated here.

[0107] In addition, as Figure 5 shown, the embodiments of the present application also provide an electronic device 500, which can be various types of computers and the like. The electronic device 500 includes a processor 510 and a memory 520. The memory 520 stores programs or instructions, which are executed by the processor 510 to implement the steps of any of the methods described above and achieve the same technical effects. To avoid repetition, details are not described here.

[0108] The embodiments of the present application also provide a readable storage medium, which stores programs or instructions, which are executed by the processor 510 to implement the steps of any of the methods described above and achieve the same technical effects. To avoid repetition, details are not described here.

[0109] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0110] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof.

[0111] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof.

[0112] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof.

[0113] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0114] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory, non-volatile memory, such as read-only memory (ROM), EPROM, and / or flash memory. The memory is an example of computer-readable media.

[0115] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0116] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or apparatus that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed or inherent to such process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0117] Those skilled in the art will appreciate that embodiments of the present application can be provided as a method, system or computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) containing computer usable program code.

[0118] The above only describes the embodiments of the present application and is not intended to limit the present application. Those skilled in the art can make various modifications and changes to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the scope of the claims of the present application.

Claims

1. A method for mining user interest, characterized in that, The method comprises: obtaining a first exposure number of a target label of a target user, the first exposure number being an exposure number of the target label in a target sliding window; determining a first weighting coefficient of the target label and a second weighting coefficient of the target label based on the first exposure number, the first weighting coefficient being a weighting coefficient for the target sliding window, and the second weighting coefficient being a weighting coefficient for historical storage; determining a positive feedback behavior index of the target label and an exposure index of the target label based on the first weighting coefficient and the second weighting coefficient, the positive feedback behavior index being an index for reflecting an overall situation of a positive feedback behavior number of the target label in the target sliding window and the historical storage, and the exposure index being an index for reflecting an overall situation of an exposure number of the target label in the target sliding window and the historical storage; determining an interest index of the target user for the target label based on the positive feedback behavior index of the target label and the exposure index of the target label; wherein the first weighting coefficient increases with an increase of the first exposure number and decreases with a decrease of the first exposure number, and the second weighting coefficient decreases with an increase of the first exposure number and increases with a decrease of the first exposure number.

2. The excavating method according to claim 1, characterized by, The mining method further comprises: obtaining a first positive feedback behavior number and a second positive feedback behavior number of a target label of a target user, the first positive feedback behavior number being a positive feedback behavior number of the target label in a target sliding window, and the second positive feedback behavior number being a positive feedback behavior number of the target label in historical storage; obtaining a second exposure number of a target label of a target user, the second exposure number being an exposure number of the target label in historical storage; determining an amplification coefficient of a target statistical quantity for the target sliding window; the determining of the positive feedback behavior index of the target label based on the first weighting coefficient and the second weighting coefficient comprises: determining the positive feedback behavior index of the target label of the target user based on the first weighting coefficient, the first positive feedback behavior number, the amplification coefficient, the second weighting coefficient, and the second positive feedback behavior number; the determining of the exposure index of the target label based on the first weighting coefficient and the second weighting coefficient comprises: determining the exposure index of the target label of the target user based on the first weighting coefficient, the first exposure number, the amplification coefficient, the second weighting coefficient, and the second exposure number, or taking the second exposure number as the exposure index of the target label of the target user.

3. The excavating method according to claim 2, characterized by, The determining of the positive feedback behavior index of the target label of the target user based on the first weighting coefficient, the first positive feedback behavior number, the amplification coefficient, the second weighting coefficient, and the second positive feedback behavior number comprises: the positive feedback behavior index of the target label of the target user is calculated by the following way: B = W1 * (N1 / R) + W2 * N2, wherein B is a positive feedback behavior index, W1 is a first weighting coefficient, N1 is a first positive feedback behavior number, R is an amplification coefficient, W2 is a second weighting coefficient, and N2 is a second positive feedback behavior number; determining, based on the first weighting coefficient, the first exposure number, the amplification coefficient, the second weighting coefficient, and the second exposure number, an exposure index of the target label of the target user, including: calculating the exposure index of the target label of the target user in the following manner: E = W1 * (M1 / R) + W2 * M2, wherein E is an exposure index, W1 is a first weighting coefficient, M1 is a first exposure number, R is an amplification coefficient, W2 is a second weighting coefficient, and M2 is a second exposure number; a sum of the second weighting coefficient and the first weighting coefficient is 1.

4. The excavating method according to claim 2, characterized by, The determining of the amplification coefficient of the statistical quantity for the target sliding window includes: taking a ratio of the first exposure number and the second exposure number as the amplification coefficient of the target statistical quantity for the target sliding window; or taking a preset value as the amplification coefficient of the target statistical quantity for the target sliding window.

5. The excavating method according to claim 1, characterized by, The target sliding window is configured to store a preset number of user behaviors for the target user in chronological order, and user behaviors exceeding the preset number are located in the historical storage; the preset number of user behaviors in the target sliding window follow a first-in-first-out principle; if a user behavior for the target label is newly added in the target sliding window, the first exposure number is increased by one statistical unit; if a user behavior for the target label is reduced in the target sliding window, the first exposure number is reduced by one statistical unit.

6. The excavating method according to any one of claims 1 to 5, characterized by, The mining method further includes: in a case where the target label meets a target condition, shielding the target label from an interest label set of the target user; in a case where the target label is in a shielding state, not pushing content corresponding to the target label to the target user.

7. The excavating method according to claim 6, characterized by, The target condition includes at least two of the following: a value of the first exposure number is greater than or equal to a set number threshold; an implicit negative feedback index of the target label of the target user in a target sliding window is greater than 0; wherein the implicit negative feedback index is an index based on indirect behavior of the target user, and is used to reflect a degree of disinterest of the target user in the target label; a click rate of the target label of the target user in the target sliding window is lower than an average click rate of the target user in the target sliding window.

8. The excavating method according to claim 7, characterized by, In a case where the target condition includes that an implicit negative feedback index of the target tag of the target user in a target sliding window is greater than 0, the implicit negative feedback index increases with an increase of user implicit negative feedback behavior for the target tag in the target sliding window, and decreases with an increase of user positive feedback behavior for the target tag in the target sliding window; the implicit negative feedback index decreases by a number of statistical units of one user positive feedback behavior, which is greater than a number of statistical units of one user implicit negative feedback behavior.

9. The excavating method according to claim 1, characterized by, After determining the interest indicator of the target user for the target tag based on the positive feedback behavior indicator of the target tag and the exposure indicator of the target tag, the mining method further includes: In a case where the interest indicator of the target tag is greater than a target threshold, adding the target tag to a set of interest tags of the target user; Pushing content corresponding to each tag in the set of interest tags to the target user.

10. A device for mining user interests, characterized in that, The apparatus includes: an obtaining module configured to obtain a first exposure frequency of a target tag of a target user, the first exposure frequency being an exposure frequency of the target tag in a target sliding window; a determining module configured to determine a first weighting coefficient of the target tag and a second weighting coefficient of the target tag based on the first exposure frequency, the first weighting coefficient being a weighting coefficient for the target sliding window, and the second weighting coefficient being a weighting coefficient for historical storage; determine a positive feedback behavior indicator of the target tag and an exposure indicator of the target tag based on the first weighting coefficient and the second weighting coefficient, the positive feedback behavior indicator being an indicator reflecting an overall situation of positive feedback behavior times of the target tag in the target sliding window and the historical storage, and the exposure indicator being an indicator reflecting an overall situation of exposure times of the target tag in the target sliding window and the historical storage; and determine an interest indicator of the target user for the target tag based on the positive feedback behavior indicator of the target tag and the exposure indicator of the target tag; wherein the first weighting coefficient increases with an increase of the first exposure frequency and decreases with a decrease of the first exposure frequency, and the second weighting coefficient decreases with an increase of the first exposure frequency and increases with a decrease of the first exposure frequency.

Images