User Label Determination Method and Device

Abstract

The present application provides a method and device for determining user tags. The method includes: obtaining the daily user behaviors of a target user within a preset period; the daily user behaviors are obtained based on the traversal of the target user among various nodes in a preset path; generating a corresponding daily behavior matrix based on the daily user behaviors; fusing and calculating to obtain a target user tag based on the daily behavior matrix and the corresponding decay weights; wherein the decay weights represent the different influences of the daily behavior matrices generated by the daily user behaviors on different dates within the preset period on the target user tag; the target user tag is used to characterize the characteristics of the target user. The method for determining user tags provided by the embodiments of the present application is used to accurately locate users based on their historical behaviors.

Description

Technical Field

[0001] This application relates to the field of intelligent human-computer interaction, and particularly to a method and device for determining user tags. Background Art

[0002] In order to improve the service quality for users and reduce the cost of manual services, before providing manual services to users, platforms have all set up intelligent assistants, also known as chatbots. The intelligent assistant can provide necessary basic services for users and solve some of the users' problems. When the intelligent assistant cannot solve the problems raised by the users, or has completed the current stage of communication and needs to turn to the next stage of communication, it will then turn to manual services.

[0003] In the related art, the dialogue state tracking (DST) technology can be used to correct the slots of the current user's message based on the user's historical chat information, so as to understand the current intention of the user and make a response based on the current intention of the user.

[0004] However, the dialogue state tracking technology can only obtain the current intention of the user, and cannot accurately locate the user, and thus cannot determine the true intention of the user's current conversation. Summary of the Invention

[0005] The purpose of this application is to provide a method and device for determining user tags, which are used to accurately locate users based on the historical behaviors of users.

[0006] This application provides a method for determining user tags, including:

[0007] Obtain the daily user behaviors of the target user within a preset period; the daily user behaviors are obtained based on the traversal between each node in the preset path by the target user; generate a corresponding daily behavior matrix based on the daily user behaviors; fuse and calculate based on the daily behavior matrix and the corresponding decay weights to obtain a target user tag; where the decay weights represent the different influences of the daily behavior matrices generated by the daily user behaviors on different dates within the preset period on the target user tag; the target user tag is used to characterize the characteristics of the target user.

[0008] Optionally, the obtaining of the daily user behaviors of the target user within a preset period includes: obtaining the target user behaviors within any target unit time within the preset period; the generating of a corresponding daily behavior matrix based on the daily user behaviors includes: generating a target behavior queue based on the target user behaviors, and generating a target behavior matrix based on the target behavior queue.

[0009] Optionally, obtaining the target user label based on the daily behavior matrix and the corresponding decay weight includes: determining the jump weight corresponding to each jump behavior according to the jump behavior between the corresponding nodes in the preset path of the target behavior queue; generating a weight matrix based on the jump weight; wherein, for the jump behavior approaching the end point of the preset path, the number of nodes spanned by each jump behavior is positively correlated with the corresponding jump weight; for the jump behavior moving away from the end point of the preset path, the number of nodes spanned by each jump behavior is negatively correlated with the corresponding jump weight.

[0010] Optionally, after generating the weight matrix based on the jump weight, the method further includes: calculating the user label to be fused of the target user within the target unit time according to the target behavior matrix and the weight matrix.

[0011] Optionally, obtaining the target user label based on the daily behavior matrix and the corresponding decay weight includes: calculating the target user label based on the user label to be fused corresponding to each time unit within the obtained preset period and the decay weight corresponding to each time unit.

[0012] Optionally, the target user label includes at least one label; the decay weight is obtained based on a decay coefficient; the jump weight is obtained based on a target coefficient; after obtaining the target user label through fusion calculation based on the daily behavior matrix and the corresponding decay weight, the method further includes: obtaining the conversation content of the target user, and determining the second target user label of the target user based on the conversation content; obtaining the optimal solutions of the decay coefficient and the target coefficient by solving the cross-entropy loss value between the target user label and the second target user label.

[0013] This application also provides a user label determination device, including:

[0014] An obtaining module, configured to obtain the daily user behavior of a target user within a preset period; the daily user behavior is obtained based on the wandering of the target user between each node in a preset path; a generating module, configured to generate a corresponding daily behavior matrix based on the daily user behavior; a calculating module, configured to obtain a target user label through fusion calculation based on the daily behavior matrix and the corresponding decay weight; wherein, the decay weight represents the different influences of the daily behavior matrices generated by the daily user behaviors on different dates within the preset period on the target user label; the target user label is used to characterize the characteristics of the target user.

[0015] Optionally, the obtaining module is specifically configured to obtain the target user behavior within any target unit time during the preset period; the generating module is specifically configured to generate a target behavior queue based on the target user behavior, and generate a target behavior matrix based on the target behavior queue.

[0016] Optionally, the apparatus further includes a determining module; the determining module is configured to determine the jump weight corresponding to each jump behavior according to the jump behavior between the corresponding nodes of the target behavior queue in the preset path; the generating module is specifically configured to generate a weight matrix based on the jump weight; wherein, for the jump behavior approaching the end point of the preset path, the number of nodes spanned by each jump behavior is positively correlated with the corresponding jump weight; for the jump behavior moving away from the end point of the preset path, the number of nodes spanned by each jump behavior is negatively correlated with the corresponding jump weight.

[0017] Optionally, the calculating module is specifically further configured to calculate the user label to be fused of the target user within the target unit time according to the target behavior matrix and the weight matrix.

[0018] Optionally, the calculating module is specifically configured to calculate the target user label based on the user label to be fused corresponding to each time unit during the obtained preset period and the attenuation weight corresponding to each time unit.

[0019] Optionally, the target user label includes at least one label; the attenuation weight is obtained based on an attenuation coefficient; the jump weight is obtained based on a target coefficient; the determining module is further configured to obtain the conversation content of the target user, and determine the second target user label of the target user based on the conversation content; the calculating module is further configured to obtain the optimal solutions of the attenuation coefficient and the target coefficient by solving the cross-entropy loss value between the target user label and the second target user label.

[0020] The present application further provides a computer program product, including computer programs / instructions, and when the computer programs / instructions are executed by a processor, the steps of the user label determination method as described in any one of the above are implemented.

[0021] The present application further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, the steps of the user label determination method as described in any one of the above are implemented.

[0022] The present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the user label determination method as described in any one of the above are implemented.

[0023] The user label determination method and device provided by this application obtain the daily user behavior of a target user based on the movement between each node in a preset path within a preset period, generate a corresponding daily behavior matrix based on the daily user behavior of the target user, and finally, based on the daily behavior matrix and the corresponding attenuation weights, infer and calculate through fusion to obtain the target user label, enabling the determination of user labels and the precise positioning of users based on the historical behavior of users in the case where relevant data is insufficient and model training cannot be used, so as to better serve users. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in this application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of this application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0025] Figure 1 is one of the flow diagrams of the user label determination method provided by this application;

[0026] Figure 2 is another flow diagram of the user label determination method provided by this application;

[0027] Figure 3 is yet another flow diagram of the user label determination method provided by this application;

[0028] Figure 4 is still another flow diagram of the user label determination method provided by this application;

[0029] Figure 5 is yet still another flow diagram of the user label determination method provided by this application;

[0030] Figure 6 is the structural diagram of the user label determination device provided by this application;

[0031] Figure 7 is the structural diagram of the electronic device provided by this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0032] To make the objectives, technical solutions, and advantages of this application clearer, the following will clearly and completely describe the technical solutions in this application in conjunction with the drawings in this application. Obviously, the described embodiments are some, but not all, of the embodiments of this application. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments in this application belong to the scope of protection of this application.

[0033] 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 interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0034] In related technologies, the dialogue state tracking (DST) technology can be used to correct the slot of the current user's message based on the user's historical chat information, so as to understand the user's current intention and respond based on the user's current intention.

[0035] Dialogue State Tracking (DST): It tracks the current state based on the domain / intention, slot-value pairs, previous state, and previous system actions. Its input is U n (the intention and slot value pair at time n, also called user action), a n-1 (system action at time n-1) and s n-1 (state at time n-1), the output is s n (state at time n).

[0036] However, conversation state tracking technology can only obtain the user's current intention, but cannot accurately locate the user, and thus cannot determine the user's true intention for the conversation.

[0037] Based on the deficiencies in the above-mentioned related technologies, the embodiments of the present application conceive of accurately locating users in the real estate field through their daily behaviors.

[0038] The following is a detailed description of the user tag determination method provided by the embodiment of the present application through specific embodiments and their application scenarios in conjunction with the accompanying drawings.

[0039] like Figure 1 As shown, an embodiment of the present application provides a method for determining a user tag, which may include the following steps 101 to 103:

[0040] Step 101: Obtain daily user behavior of a target user within a preset period.

[0041] Among them, the daily user behavior is obtained based on the movement of the target user between each node in the preset path.

[0042] Exemplarily, for the real estate field, the browsing behavior of users on the platform may include: searching, sorting, clicking, detail page, apartment type page, following, etc. Users can perform at least one of the above behaviors on the platform.

[0043] Exemplarily, based on the browsing behavior of users on the platform, a preset path of user behavior can be obtained. The behavior of users based on this preset path will ultimately end with the behavior of entering instant messaging (IM). That is, the final behavior path of users will enter IM, and entering IM belongs to a complete user behavior path.

[0044] Exemplarily, users can start from any node in this preset path and end with entering IM. At the same time, users can move around in this path, that is, users can jump from one node to any other node.

[0045] For example, as Figure 2 shown, it is an abstraction of the user browsing path, from left to right are: searching, sorting, clicking, detail page, apartment type page, following. That is, when a certain user searches for housing sources on the platform, they can search for the intended housing sources in the above path order.

[0046] At the same time, users can also directly click on the housing sources recommended on the home page and directly follow the housing sources when the housing sources are the intended ones, that is, users can randomly jump between any nodes in this path.

[0047] Exemplarily, the daily user behavior of the target user can be obtained according to the browsing logs of the user. In order to obtain the above daily user behavior, it is necessary to obtain the daily browsing logs of the user within a preset period and obtain the daily behavior of the user according to the daily browsing logs of the user.

[0048] Specifically, the above preset period can be T days, and the daily behavior of the user can be obtained according to the browsing logs of the user every day within T days. In the embodiments of the present disclosure, the obtaining method of the daily behavior or browsing logs of the user is not limited. For example, it can be obtained by recording the behavior of the user with the authorization of the user to obtain the daily behavior or browsing logs of the user.

[0049] It can be understood that since the needs of users are relatively stable within a period of time, the label of the user can be inferred by obtaining the daily behavior of the user in the recent period of time, and then the accurate positioning of the user can be completed.

[0050] Step 102: Generate a corresponding daily behavior matrix based on the daily user behavior.

[0051] Exemplarily, after obtaining the daily user behavior of the above target user, a corresponding daily behavior matrix can be generated based on the daily user behavior.

[0052] Specifically, the above preset period may include: t1, t2, t3... T, that is, T days with days as the unit. The obtained daily user behavior includes: s1, s2, s3... sT, and the corresponding daily behavior matrices are M1, M2, M3... MT. Among them, according to the user behavior obtained within any unit time, multiple behavior matrices can be generated. That is, the above Mn may include multiple behavior matrices, n ∈ [1, T].

[0053] For example, as Figure 3 shown, by obtaining the browsing records of the user for three days, according to the user's daily browsing log (for example, Day1 log, indicating the browsing log of the user on the first day within the preset period), the daily behavior queue of the user can be obtained. Then, based on the daily behavior queue, the corresponding daily behavior matrix can be obtained, and there is a corresponding behavior matrix for each day. This behavior matrix is used to reflect the behavior of the user, where 0 indicates that the user did not perform this behavior, and 1 indicates that the user performed this behavior. Taking the preset path including 4 nodes, one node corresponding to one behavior, and the user's browsing behavior including: node 1, node 3, node 4, node 1 as an example, the behavior matrix 1 obtained based on the browsing log Day1 log of the user on the first day within the preset period:

[0054]

[0055] The first row 1 0 1 0 in this matrix 1 indicates that the user executed the behavior corresponding to node 1 and jumped from node 1 to node 3; the second row 0 0 0 1 in this matrix 1 indicates that the user jumped from node 3 to node 4; the fourth row 1 0 0 1 in this matrix 1 indicates that the user jumped back from node 4 to node 1. Thus, a corresponding daily node matrix can be generated based on the daily nodes of the user.

[0056] Step 103: Based on the daily behavior matrix and the corresponding decay weight, calculate through fusion to obtain the target user label.

[0057] Among them, the decay weight represents the different influences of the daily behavior matrices generated by the daily user behavior on different dates within the preset period on the target user label. The target user label is used to characterize the characteristics of the target user.

[0058] It should be noted that the above target user label may include at least one label of the target user, that is, according to the daily browsing behavior of the user, at least one label of the user can be inferred.

[0059] Exemplarily, after obtaining the daily behavior matrix of the user based on the user's roaming behavior in the above-mentioned preset path, the target user label can be obtained through fusion calculation according to the daily behavior matrix and the corresponding decay weight. This roaming behavior can be random or executed according to a certain rule.

[0060] It can be understood that since within the above-mentioned preset period, the user behavior closer to the current time can better reflect the user's current intention, therefore, the decay weight can be used to adjust the influence of the daily behavior matrix on the finally obtained user label.

[0061] Exemplarily, the above-mentioned decay weight can affect the jump weight in units of days based on the following formula:

[0062] W = e (-r*day) (Formula 1)

[0063] Wherein, W represents the decay weight of the influence of the daily behavior matrix on the finally obtained user label. This jump weight increases with the increase of the number of days day, and r represents the decay coefficient. Day represents the day of the preset period. The user behavior earlier from the current time has less influence on the final target user label.

[0064] Optionally, in the embodiments of the present application, the user behavior within the unit time in units of days within the preset period can be specifically obtained through the following steps, and the corresponding behavior matrix can be generated according to the user behavior.

[0065] Exemplarily, the above-mentioned step 101 may include the following step 101a:

[0066] Step 101a: Obtain the target user behavior within any target unit time within the preset period.

[0067] Exemplarily, any of the above-mentioned target unit times is any day within the above-mentioned preset period.

[0068] Exemplarily, after obtaining the user behavior on any day within the preset period, the above-mentioned step 102 may include the following step 102a:

[0069] Step 102a: Generate a target behavior queue based on the target user behavior, and generate a target behavior matrix based on the target behavior queue.

[0070] Exemplarily, the generation steps of the above-mentioned target behavior matrix can refer to the specific generation steps of generating the user's daily behavior matrix based on the user's daily browsing behavior as shown in Figure 3 the following.

[0071] It should be noted that for the user behavior matrix of each day within the above-mentioned preset period, it can be obtained according to the methods in step 101a and step 102a above.

[0072] Exemplarily, after obtaining the target behavior matrix of the target user within any target unit time, it is also necessary to consider the jump weight of each jump behavior when the user wanders in the above-mentioned preset path.

[0073] Exemplarily, the above step 103 may include the following step 103a1 and step 103a2:

[0074] Step 103a1: Determine the jump weight corresponding to each jump behavior according to the jump behavior between the corresponding nodes in the preset path of the target behavior queue.

[0075] Step 103a2: Generate a weight matrix based on the jump weight.

[0076] Among them, for the jump behavior approaching the end point of the preset path, the number of nodes spanned by each jump behavior is positively correlated with the corresponding jump weight; for the jump behavior moving away from the end point of the preset path, the number of nodes spanned by each jump behavior is negatively correlated with the corresponding jump weight.

[0077] Exemplarily, as Figure 2 shown, for the wandering behavior of the user in the above-mentioned preset path, the weight W of the jump behavior closer to the end point is larger. At the same time, for the jump behavior of the user's retreat, it is necessary to reduce the jump weight of this jump behavior to a certain extent.

[0078] Specifically, the following formula can be referred to calculate the jump weight of each jump behavior when the user wanders in the above-mentioned preset path:

[0079]

[0080] Among them, W i,j represents the weight of the user jumping from behavior i to behavior j; λ is a coefficient, and ind represents the index corresponding to the user behavior.

[0081] Optionally, in the embodiment of the present application, after obtaining the above-mentioned target user label, the parameters r in the above formula one and the parameter λ in the formula two can be solved based on the target user label.

[0082] Exemplarily, the target user label includes at least one label; the decay weight is obtained based on the decay coefficient (i.e., the parameter r in the above formula one); the jump weight is obtained based on the target coefficient (i.e., λ in the above formula two).

[0083] Exemplarily, after the above step 103, the user label determination method provided by the embodiments of the present application may further include the following steps 104 and 105:

[0084] Step 104: Obtain the conversation content of the target user, and determine the second target user label of the target user based on the conversation content.

[0085] Step 105: Obtain the optimal solutions of the attenuation coefficient and the target coefficient by solving the cross-entropy loss value between the target user label and the second target user label.

[0086] Exemplarily, the above target user label in the embodiments of the present application is calculated based on the following formula three:

[0087]

[0088] Among them, infer represents the user label obtained by inference, that is, the above target user label; pic is used to indicate the number of behavior matrices obtained based on the browsing behavior of the user, pi represents the i-th behavior matrix; Aq represents the action queue obtained based on the browsing behavior of the user; obj represents the queue generated based on the daily browsing log of the user.

[0089] Exemplarily, since the target user label includes at least one label, therefore, the above infer may be the label distribution of the user. The optimal values of r in the above formula one and λ in the above formula two are solved through this user label distribution and the following formula four:

[0090]

[0091] Among them, infer(i) identifies the i-th target user label in the target user label distribution; tager(i) represents the i-th label in the user label distribution obtained based on the conversation content of the user.

[0092] Exemplarily, based on the loss function in the above formula four, r in the above formula one and λ in the above formula two can be optimized and solved through gradient descent, and finally the optimal fitting parameters are obtained.

[0093] Exemplarily, the index corresponding to the user behavior may refer to Table 1 below:

[0094] Behavior Search Sort Click Detail page House type page Follow Index 1 2 3 4 5 6

[0095] Table 1

[0096] Exemplarily, for the above behavior matrix, above the diagonal (i.e., Figure 3 the bold part of the behavior matrix in Figure 2The part considered when walking in the path sequence shown. Below the diagonal is the part considered when the user retreats in the path sequence as shown in Figure 2 .

[0097] For example, as shown in Figure 3 , according to each jump behavior of the user, the jump weight corresponding to the walking behavior of the user in the preset path can be obtained, that is, each element of the behavior matrix has a corresponding weight.

[0098] Exemplarily, after obtaining the behavior matrix and the corresponding weight matrix of the user within any unit time, the user label corresponding to the any unit time can be calculated.

[0099] Exemplarily, after the above step 103a2, the user label determination method provided by the embodiment of the present application may further include the following step 103a3:

[0100] Step 103a3: Calculate the user label to be fused of the target user within the target unit time according to the target behavior matrix and the weight matrix.

[0101] It can be understood that for the user behavior obtained on any day within the above preset period, the user label of that day can be deduced, that is, based on the user's behavior on that day, the user label of that day can be obtained. Then, according to the obtained multiple user labels, the comprehensive label of the user within the preset period is obtained after fusion calculation.

[0102] For example, based on Figure 3 , as shown in Figure 4 , after obtaining the behavior matrix and the jump weight corresponding to the browsing behavior of the user every day within three days, according to the behavior matrix and the corresponding jump weight of each day, the user persona corresponding to each day is calculated, that is, the user label corresponding to each unit time of the target user within the preset period.

[0103] Since the behavior of one day cannot fully reflect the user's intention and thus cannot accurately locate the user, it is necessary to obtain multiple samples, that is, multiple user labels, according to the methods in the above steps 103a1 to 103a3, and then perform fusion calculation on the multiple user labels to obtain the target user label that can accurately locate the user.

[0104] Exemplarily, the above step 103 may include the following step 103b:

[0105] Step 103b: Calculate the target user label based on the user label to be fused corresponding to each time unit within the obtained preset period and the decay weight corresponding to each time unit.

[0106] Exemplarily, the to-be-fused user tags corresponding to each of the above time units are obtained according to the methods in the above steps 103a1 to 103a3.

[0107] For example, combining Figure 3 and Figure 4 ,like Figure 5 As shown in the figure, after obtaining the user tags corresponding to the browsing behavior of the user every day in three days, the user tags that can accurately locate the user can be obtained through comprehensive calculation based on the decay weight corresponding to each day, that is, the above-mentioned target user tags. Then, based on the target user tags, targeted replies can be made to the user's conversation on the smart assistant.

[0108] The user label determination method provided in the embodiment of the present application obtains the daily user behavior of the target user based on the target user's wandering between nodes in a preset path within a preset period, and generates a corresponding daily behavior matrix based on the target user's daily user behavior. Finally, based on the daily behavior matrix and the corresponding attenuation weight, the target user label is obtained after reasoning and fusion calculation. In the case where there is insufficient relevant data and model training cannot be used, the user label can be determined based on the user's historical behavior and the user can be accurately located to better serve the user.

[0109] It should be noted that the user tag determination method provided in the embodiment of the present application can be executed by a user tag determination device, or a control module in the user tag determination device for executing the user tag determination method. In the embodiment of the present application, the user tag determination device provided in the embodiment of the present application is described by taking the user tag determination device executing the user tag determination method as an example.

[0110] It should be noted that in the embodiments of the present application, the user tag determination methods shown in the above-mentioned method drawings are all illustrated by combining one of the drawings in the embodiments of the present application as an example. In specific implementation, the user tag determination methods shown in the above-mentioned method drawings can also be implemented in combination with any other drawings that can be combined as shown in the above-mentioned embodiments, which will not be repeated here.

[0111] The user tag determination device provided in the present application is described below, and the user tag determination method described below and described above can be referenced to each other.

[0112] Figure 6 A schematic diagram of the structure of a user tag determination device provided in an embodiment of the present application is shown as follows: Figure 6 As shown, specifically including:

[0113] An acquisition module 601, configured to acquire the daily user behavior of a target user within a preset period; the daily user behavior is obtained based on the movement of the target user between each node in a preset path; a generation module 602, configured to generate a corresponding daily behavior matrix based on the daily user behavior; a calculation module 603, configured to fuse and calculate based on the daily behavior matrix and the corresponding decay weight to obtain a target user label; wherein, the decay weight represents the different influences of the daily behavior matrices generated by the daily user behavior on different dates within the preset period on the target user label; the target user label is used to characterize the characteristics of the target user.

[0114] Optionally, the acquisition module 601 is specifically configured to acquire the target user behavior within any target unit time within the preset period; the generation module 602 is specifically configured to generate a target behavior queue based on the target user behavior, and generate a target behavior matrix based on the target behavior queue.

[0115] Optionally, the apparatus further includes: a determination module 604; the determination module 604 is configured to determine a jump weight corresponding to each jump behavior according to the jump behavior of the target behavior queue between the corresponding nodes in the preset path; the generation module 602 is specifically configured to generate a weight matrix based on the jump weight; wherein, for the jump behavior approaching the end point of the preset path, the number of nodes spanned by each jump behavior is positively correlated with the corresponding jump weight; for the jump behavior moving away from the end point of the preset path, the number of nodes spanned by each jump behavior is negatively correlated with the corresponding jump weight.

[0116] Optionally, the calculation module 603 is specifically further configured to calculate the user label to be fused of the target user within the target unit time according to the target behavior matrix and the weight matrix.

[0117] Optionally, the calculation module 603 is specifically configured to calculate the target user label based on the user label to be fused corresponding to each time unit within the acquired preset period and the decay weight corresponding to each time unit.

[0118] Optionally, the target user label includes at least one label; the decay weight is obtained based on a decay coefficient; the jump weight is obtained based on a target coefficient; the determination module is further configured to acquire the conversation content of the target user, and determine a second target user label of the target user based on the conversation content; the calculation module is further configured to obtain the optimal solutions of the decay coefficient and the target coefficient by solving the cross-entropy loss value between the target user label and the second target user label.

[0119] The user label determination device provided by this application obtains the daily user behaviors of a target user within a preset period based on the traversal between nodes in a preset path by the target user, generates a corresponding daily behavior matrix based on the daily user behaviors of the target user, and finally, through inference and fusion calculation based on the daily behavior matrix and the corresponding decay weights, obtains the target user label, enabling the determination of user labels and the precise positioning of users based on the historical behaviors of users in the case where relevant data is insufficient and model training cannot be used, so as to better serve users.

[0120] Figure 7 An example of the physical structure diagram of an electronic device is as Figure 7 shown. The electronic device may include: a processor 710, a communication interface 720, a memory 730, and a communication bus 740. Among them, the processor 710, the communication interface 720, and the memory 730 communicate with each other through the communication bus 740. The processor 710 can call the logical instructions in the memory 730 to execute the user label determination method, which includes: obtaining the daily user behaviors of a target user within a preset period; the daily user behaviors are obtained based on the traversal between nodes in a preset path by the target user; generating a corresponding daily behavior matrix based on the daily user behaviors; through inference and fusion calculation based on the daily behavior matrix and the corresponding decay weights, obtaining the target user label; where the decay weight represents the different influences of the daily behavior matrices generated by the daily user behaviors on different dates within the preset period on the target user label; the target user label is used to characterize the characteristics of the target user.

[0121] In addition, when the logical instructions in the above-mentioned memory 730 are implemented in the form of software functional units and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0122] On the other hand, the present application also provides a computer program product, which includes a computer program stored on a computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the user label determination method provided by each of the above methods. The method includes: obtaining the daily user behavior of a target user within a preset period; the daily user behavior is obtained based on the movement of the target user between each node in a preset path; generating a corresponding daily behavior matrix based on the daily user behavior; obtaining a target user label through fusion calculation based on the daily behavior matrix and the corresponding attenuation weight; wherein, the attenuation weight represents the different influences of the daily behavior matrices generated by the daily user behavior on different dates within the preset period on the target user label; the target user label is used to characterize the characteristics of the target user.

[0123] In another aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the user label determination method provided by each of the above. The method includes: obtaining the daily user behavior of a target user within a preset period; the daily user behavior is obtained based on the movement of the target user between each node in a preset path; generating a corresponding daily behavior matrix based on the daily user behavior; obtaining a target user label through fusion calculation based on the daily behavior matrix and the corresponding attenuation weight; wherein, the attenuation weight represents the different influences of the daily behavior matrices generated by the daily user behavior on different dates within the preset period on the target user label; the target user label is used to characterize the characteristics of the target user.

[0124] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0125] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for determining user tags, characterized in that, Including: Obtain the daily user behavior of the target user within a preset period; The daily user behavior is obtained based on the movement of the target user between each node in the preset path; Based on the daily user behavior, generate a corresponding daily behavior matrix; Based on the daily behavior matrix and the corresponding decay weight, obtain the target user label after fusion calculation; The step of obtaining the target user label after fusion calculation based on the daily behavior matrix and the corresponding decay weight specifically includes: According to the jump behavior of the target behavior queue between the corresponding nodes in the preset path, determine the jump weight corresponding to each jump behavior, and generate a weight matrix based on the jump weight; According to the target behavior matrix and the weight matrix, calculate the user label to be fused of the target user within the target unit time; Based on the user label to be fused corresponding to each time unit within the obtained preset period, and the decay weight corresponding to each time unit, calculate the target user label; Wherein, the target behavior queue is generated based on the target user behavior corresponding to the target user, and the target user behavior is the daily user behavior corresponding to the target user within any target unit time in the preset period; the target behavior matrix is generated based on the target behavior queue; the decay weight represents the different influences of the daily behavior matrices generated by the daily user behaviors on different dates within the preset period on the target user label; the target user label is used to characterize the characteristics of the target user.

2. The method according to claim 1, wherein The method further includes: For the jump behavior approaching the end point of the preset path, the number of nodes spanned by each jump behavior is positively correlated with the corresponding jump weight; for the jump behavior moving away from the end point of the preset path, the number of nodes spanned by each jump behavior is negatively correlated with the corresponding jump weight.

3. The method according to claim 1, wherein The target user label includes at least one label; the decay weight is obtained based on a decay coefficient; the jump weight is obtained based on a target coefficient; After obtaining the target user label through fusion calculation based on the daily behavior matrix and the corresponding decay weight, the method further includes: Obtain the conversation content of the target user, and determine the second target user label of the target user based on the conversation content; By solving the cross-entropy loss value between the target user label and the second target user label, obtain the optimal solutions of the decay coefficient and the target coefficient.

4. A user label determination device, characterized in that, The device includes: An acquisition module, configured to acquire the daily user behavior of the target user within a preset period; the daily user behavior is obtained based on the movement of the target user between each node in the preset path; A generation module, configured to generate a corresponding daily behavior matrix based on the daily user behavior; A calculation module, configured to obtain the target user label after fusion calculation based on the daily behavior matrix and the corresponding decay weight; the step of obtaining the target user label after fusion calculation based on the daily behavior matrix and the corresponding decay weight is specifically used for: Determine the jump weight corresponding to each jump behavior according to the jump behavior between the corresponding nodes in the preset path for the target behavior queue, and generate a weight matrix based on the jump weight; Calculate the user tags to be fused for the target user within the target unit time according to the target behavior matrix and the weight matrix; Calculate the target user tag based on the user tags to be fused corresponding to each time unit within the obtained preset period and the decay weight corresponding to each time unit; Wherein, the target behavior queue is generated based on the target user behavior corresponding to the target user, and the target user behavior is the daily user behavior corresponding to the target user within any target unit time within the preset period; the target behavior matrix is generated based on the target behavior queue; the decay weight represents the different influences of the daily behavior matrices generated by the daily user behaviors on different dates within the preset period on the target user tag; the target user tag is used to characterize the characteristics of the target user.

5. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the user tag determination method according to any one of claims 1 to 3.

6. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, it implements the steps of the user tag determination method according to any one of claims 1 to 3.

Images