A method, device, medium and equipment for obtaining user portrait prediction model

Abstract

The present invention provides a method, device, medium and equipment for obtaining a user portrait prediction model, and relates to the field of data processing technology, including: obtaining an initial prediction model; obtaining a specified user ID list and a specified feature vector list set corresponding to the specified user ID list; inputting the specified feature vector list set into the initial prediction model to obtain a target priority; continuously adjusting the parameters corresponding to the initial prediction model until the target priority meets a preset priority threshold to obtain a target user portrait prediction model; it can be seen that a large language model is used to process feature information corresponding to the user, which solves the problem of large and complex data volume and improves the efficiency of predicting user portraits. When training the large language model, the model prediction value is not directly compared with the true value, and the filtered feature information is compared through feature screening, thereby improving the accuracy of obtaining the target user portrait prediction model.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, device, medium and equipment for obtaining a user portrait prediction model. Background Art

[0002] User portrait data has important guiding significance for data analysis, such as the analysis of user behavior data. It outlines the user's various personal characteristics, and then explores user needs and analyzes user preferences. By matching user portraits, it can provide users with more efficient and targeted information delivery and a user experience that is closer to personal habits. However, the feature information data included in user portraits is diverse and complex, and there are problems such as the inability to directly obtain comprehensive user portraits. How to accurately use existing information to predict user portraits has become an important focus and research direction at present.

[0003] The method for obtaining user portrait prediction models in the prior art is that technical specialists summarize and generalize the long-term characteristics of customer groups based on business experience and use this to describe and predict user portraits to build user portrait prediction models. This method relies on personal experience, consumes manpower, and cannot capture all aspects of user information changes, thereby reducing the accuracy of predicting user portraits. Summary of the invention

[0004] In view of the above technical problems, the technical solution adopted by the present invention is a method for obtaining a user portrait prediction model, and the method comprises the following steps:

[0005] Get the initial prediction model.

[0006] Obtaining a designated user ID list and a designated feature vector list set corresponding to the designated user ID list, wherein the designated user ID list includes a plurality of designated user IDs, and obtaining the designated user ID list and a designated feature vector list set corresponding to the designated user ID list includes the following steps:

[0007] A first designated data list corresponding to each designated user ID is obtained, wherein the first designated data list includes a plurality of first designated data, wherein the first designated data is data related to initial features involved in a designated user corresponding to the designated user ID within a preset time period.

[0008] According to the first specified data list, a first specified feature vector is obtained.

[0009] A second designated data list corresponding to each designated user ID is obtained, wherein the second designated data list includes a plurality of second designated data, wherein the second designated data is data related to the initial features involved by the designated user corresponding to the designated user ID within a designated time period.

[0010] According to the second specified data list, a second specified feature vector is obtained.

[0011] A designated feature vector is obtained according to the first designated feature vector and the second designated feature vector.

[0012] Input the specified feature vector list set into the initial prediction model to obtain the target priority.

[0013] The parameters corresponding to the initial prediction model are continuously adjusted until the target priority meets the preset priority threshold to obtain the target user portrait prediction model.

[0014] The present invention also provides a device for obtaining a user portrait prediction model, the device comprising:

[0015] The first processing module is used to obtain an initial prediction model.

[0016] A second processing module is used to obtain a specified user ID list and a specified feature vector list set corresponding to the specified user ID list, wherein the second processing module also includes.

[0017] The first designated data list acquisition module is used to obtain the first designated data list corresponding to each designated user ID, wherein the first designated data list includes a plurality of first designated data, wherein the first designated data is data related to the initial features involved in the designated user corresponding to the designated user ID within a preset time period.

[0018] The first designated feature vector acquisition module is used to acquire the first designated feature vector according to the first designated data list.

[0019] The second designated data list acquisition module is used to obtain the second designated data list corresponding to each designated user ID, wherein the second designated data list includes a plurality of second designated data, wherein the second designated data is data related to the initial features involved in the designated user corresponding to the designated user ID within a designated time period.

[0020] A second designated feature vector acquisition module, used for acquiring a second designated feature vector according to a second designated data list;

[0021] The designated feature vector acquisition module is used to acquire the designated feature vector according to the first designated feature vector and the second designated feature vector.

[0022] The third processing module is used to input the specified feature vector list set into the initial prediction model to obtain the target priority.

[0023] The fourth processing module is used to continuously adjust the parameters corresponding to the initial prediction model until the target priority meets the preset priority threshold to obtain the target user portrait prediction model.

[0024] The present invention also provides a non-transitory computer-readable storage medium, in which at least one instruction or at least one program is stored, and the at least one instruction or at least one program is loaded and executed by a processor to implement the above-mentioned method of obtaining a user portrait prediction model.

[0025] The present invention also provides an electronic device, comprising a processor and the above-mentioned non-transitory computer-readable storage medium.

[0026] The present invention has at least the following beneficial effects: obtaining an initial prediction model; obtaining a specified user ID list and a specified feature vector list set corresponding to the specified user ID list; inputting the specified feature vector list set into the initial prediction model to obtain a target priority; continuously adjusting the parameters corresponding to the initial prediction model until the target priority meets a preset priority threshold to obtain a target user portrait prediction model; it can be seen that the use of a large language model to process the feature information corresponding to the user solves the problem of large and complex data volume and improves the efficiency of predicting user portraits. When training the large language model, the model prediction value is not directly compared with the true value, and the filtered feature information is compared through feature screening, thereby improving the accuracy of obtaining the target user portrait prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0028] Figure 1 A flow chart of a method for obtaining a user portrait prediction model provided in the first embodiment of the present invention;

[0029] Figure 2 A flowchart of S2 provided in Embodiment 1 of the present invention;

[0030] Figure 3 This is a flow chart of S3 provided in Embodiment 1 of the present invention;

[0031] Figure 4 This is a flow chart of S33 provided in the first embodiment of the present invention;

[0032] Figure 5 A schematic diagram of the structure of a device for obtaining a user portrait prediction model provided in Embodiment 2 of the present invention;

[0033] Figure 6 A schematic diagram of the structure of a second execution module 2 provided in Embodiment 2 of the present invention;

[0034] Figure 7 A schematic diagram of the structure of the third execution module 3 provided in the second embodiment of the present invention;

[0035] Figure 8 This is a schematic diagram of the structure of the target acquisition module 33 provided in the second embodiment of the present invention. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0037] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0038] Embodiment 1

[0039] This embodiment provides a method for obtaining a user portrait prediction model, the method comprising the following steps: Figure 1 As shown:

[0040] S1, obtain the initial prediction model.

[0041] Specifically, the initial prediction model is a prediction model obtained by inputting the initial user ID list and the target feature vector list set into the initial LLM model for training.

[0042] Furthermore, those skilled in the art know that the methods for training LLM models in the prior art all fall within the protection scope of the present invention and will not be described in detail herein.

[0043] Specifically, the initial user ID list includes a plurality of initial user IDs, wherein the initial user ID is a unique identifier representing an initial user.

[0044] Furthermore, the initial user is a user of a device with a communication function used for model training.

[0045] Furthermore, those skilled in the art know that any method for obtaining a user identifier in the prior art falls within the protection scope of the present invention and will not be described in detail herein.

[0046] Specifically, the target feature vector list set includes several target feature vector lists, wherein each initial user ID corresponds to one target feature vector list, and the target feature vector list includes several target feature vectors.

[0047] Specifically, S1 also includes the following steps:

[0048] S101, obtaining an initial feature data list corresponding to each initial user ID.

[0049] Specifically, the initial feature data list includes a plurality of initial feature data, wherein the initial feature data is data related to the initial feature that the initial user corresponding to the initial user ID is involved in before the current moment.

[0050] Furthermore, the initial features are the types of data related to the initial user obtained through the device of the initial user, such as: a list of installed apps, the names of installed apps and the time of installation, the names of uninstalled apps and the time of uninstallation, a list of used apps, geographic location information that has appeared, descriptions corresponding to installed apps, descriptions corresponding to uninstalled apps, and other initial features.

[0051] S102: Acquire a first data list and a second data list corresponding to the initial user ID according to the initial feature data list.

[0052] Specifically, the first data list includes a plurality of first data, wherein the first data is sequence data and the like.

[0053] Further, when the initial feature data in the initial feature data list can be converted into a sequence, the sequence corresponding to the initial feature data obtained is the first data, for example: the name of the installed APP and the time of installation.

[0054] Specifically, the second data list includes a plurality of second data, wherein the second data is text data.

[0055] Further, when the initial feature data in the initial feature data list is text, the acquired text corresponding to the initial feature data is the second data, for example: a description of an APP, a message pushed to a user device, and the like.

[0056] S103: Obtain a target feature vector list corresponding to each initial user ID according to the first data list and the second data list.

[0057] Specifically, the target feature vector list includes several target feature vectors.

[0058] Specifically, S103 also includes the following steps:

[0059] S1031, obtaining a first target feature vector according to the first data list.

[0060] Specifically, the first target feature vector is a vector obtained by converting the first data in the first data list into a vector and then adding them together. It can be understood as: obtaining the i-th first data vector A i =(A i1 , ..., A ij , ..., A im ), A ij is the jth bit in the i-th first data vector, the value range of j is 1 to m, m is the dimension of the first data vector, and the first target feature vector B=(B1, ..., B j , ..., B m ), where B j =∑ n i=1 A ij , the value range of i is 1 to n, and n is the number of first data in the first data list.

[0061] Furthermore, those skilled in the art know that any method for converting a sequence into a vector in the prior art falls within the protection scope of the present invention and will not be described in detail herein.

[0062] S1032: Obtain a second target feature vector according to the second data list.

[0063] Specifically, the second target feature vector is a vector obtained by converting the second data in the second data list into vectors and then adding them.

[0064] Furthermore, the second target feature vector is obtained in the same manner as the first target feature vector.

[0065] S1033: Obtain a target feature vector according to the first target feature vector and the second target feature vector.

[0066] Specifically, the target feature vector is a vector obtained by adding the first target feature vector and the second target feature vector.

[0067] Furthermore, the method for acquiring the target feature vector is consistent with the method for acquiring the first target feature vector.

[0068] Based on the above content, we can know that by uniformly converting different types of feature data corresponding to users into vectors for training the large language model, we can avoid the problem of non-standardized data caused by diverse sources and different formats of user data. By understanding the corresponding semantic relationships through the large language model, we can obtain the correlation between data and data, thus improving the accuracy of obtaining the target user portrait.

[0069] S2, obtaining a specified user ID list and a specified feature vector list set corresponding to the specified user ID list.

[0070] Specifically, the designated user ID list includes a plurality of designated user IDs, wherein the designated user ID is a unique identifier representing a designated user.

[0071] Furthermore, the designated user is a user having a communication device for detecting the initial prediction model.

[0072] Furthermore, the method for obtaining the designated user ID is consistent with the method for obtaining the initial user ID.

[0073] Specifically, the designated feature vector list set includes a plurality of designated feature vector lists, wherein each designated user ID included in the designated user ID list corresponds to a designated feature vector list, and the designated feature vector list includes a plurality of designated feature vectors.

[0074] Specifically, S2 includes the following steps: Figure 2 As shown:

[0075] S21, obtaining a first designated data list corresponding to each designated user ID.

[0076] Specifically, the first designated data list includes a plurality of first designated data, wherein the first designated data is data related to the initial features and involved in a designated user corresponding to a designated user ID within a preset time period.

[0077] Furthermore, the end time of the preset time period is before the current time.

[0078] Furthermore, those skilled in the art know that the time span corresponding to the preset time period can be selected according to actual needs, which falls within the protection scope of the present invention and will not be described in detail here.

[0079] S22, obtaining a first designated feature vector according to the first designated data list.

[0080] Specifically, the method for acquiring the first designated feature vector is consistent with the method for acquiring the target feature vector, and may refer to steps S102 to S103.

[0081] S23, obtaining a second designated data list corresponding to each designated user ID.

[0082] Specifically, the second designated data list includes a plurality of second designated data, wherein the second designated data is data related to the initial features and involved in a designated user corresponding to a designated user ID within a designated time period.

[0083] Furthermore, the start time of the specified time period is the end time of the preset time period.

[0084] Furthermore, the end time of the specified time period is the current time.

[0085] S24, obtaining a second designated feature vector according to the second designated data list.

[0086] Specifically, the method for acquiring the second designated feature vector is consistent with the method for acquiring the target feature vector, and reference may be made to steps S102 to S103.

[0087] S25, obtaining a designated feature vector according to the first designated feature vector and the second designated feature vector.

[0088] Specifically, the designated feature vector includes a first designated feature vector and a second designated feature vector.

[0089] S3, input the specified feature vector list set into the initial prediction model to obtain the target priority.

[0090] Specifically, S3 also includes the following steps: Figure 3 As shown:

[0091] S31, obtaining a first intermediate feature vector list corresponding to a specified user ID list.

[0092] Specifically, the first intermediate feature vector list includes a plurality of first intermediate feature vectors, wherein each designated user ID corresponds to a first intermediate feature vector.

[0093] Furthermore, the first intermediate feature vector is a feature vector corresponding to a specified time period obtained by inputting the first specified feature vector into an initial prediction model for prediction.

[0094] S32, obtaining a second intermediate feature vector list corresponding to the specified user ID list.

[0095] Specifically, the second intermediate feature vector list includes a plurality of second intermediate feature vectors, wherein each designated user ID corresponds to a second intermediate feature vector.

[0096] Furthermore, the second intermediate feature vector is a second designated feature vector corresponding to the acquired designated user ID.

[0097] S33: Obtain target priority according to the first intermediate feature vector list and the second intermediate feature vector list.

[0098] Specifically, S33 also includes the following steps: Figure 4 As shown:

[0099] S331, according to the first intermediate feature vector list, obtain a first key feature vector list, that is, according to the first intermediate feature vector list, obtain a first key feature vector list H={H1, ..., H g , ..., H h}, H g is the gth first key eigenvector, the value range of g is 1 to h, and h is the number of first key eigenvectors.

[0100] Specifically, the first key feature vector list includes a first key feature vector, wherein the first key feature vector is a feature vector corresponding to a specific feature obtained from the first intermediate feature vector.

[0101] Furthermore, the specific feature is an initial feature in which an initial feature corresponding to the first intermediate feature vector overlaps with an initial feature corresponding to the second intermediate feature vector.

[0102] S332, according to the second intermediate feature vector list, obtain a second key feature vector list, that is, according to the second intermediate feature vector list, obtain a second key feature vector list P={P1, ..., P g , ..., P h}, P g is the g-th second key eigenvector.

[0103] Specifically, the second key feature vector list includes a second key feature vector, wherein the second key feature vector is a feature vector corresponding to a specific feature obtained from the second intermediate feature vector.

[0104] S333: Obtain a target priority according to the first key feature vector list and the second key feature vector list, that is, obtain a target priority F according to H and P, where F meets the following conditions:

[0105] F=∑ hg=1 Q g , Q g H g With P g The similarity between .

[0106] Specifically, those skilled in the art know that any method for calculating the similarity between vectors in the prior art falls within the protection scope of the present invention and will not be described in detail herein.

[0107] S4, continuously adjust the parameters corresponding to the initial prediction model until the target priority meets the preset priority threshold to obtain the target user portrait prediction model.

[0108] Specifically, the preset priority threshold value ranges from 0.75 to 0.9, wherein those skilled in the art know that the preset priority threshold value can be selected according to actual needs, all of which fall within the protection scope of the present invention and will not be described in detail here.

[0109] From the above content, it can be seen that the use of a large language model to process the feature information corresponding to the user solves the problem of large and complex data volume and improves the efficiency of predicting user portraits. When training the large language model, the model prediction value is not directly compared with the true value. The feature information after feature screening is compared through feature screening, which improves the accuracy of the target user portrait prediction model.

[0110] In another specific embodiment, after S1033, the following steps are also included:

[0111] S200, constructing a target training model according to the target feature vector list set.

[0112] Specifically, S200 also includes the following steps:

[0113] S201, inputting the target feature vector list set into the initial LLM model for training.

[0114] S202, continuously adjusting the parameters of the initial LLM model and screening the initial features to adjust the target feature vector list set until the error is less than a preset threshold to obtain a target training model.

[0115] Specifically, the preset threshold value ranges from 0.2 to 0.3, wherein those skilled in the art know that the preset threshold value can be selected according to actual needs, all of which fall within the protection scope of the present invention and will not be described in detail here.

[0116] Specifically, those skilled in the art know that any method of adjusting model parameters and screening features to change model input in the prior art falls within the protection scope of the present invention and will not be described in detail herein.

[0117] S300, obtaining a target feature data list corresponding to a target user ID.

[0118] Specifically, the target user ID is a unique identifier that represents the target user, wherein the method for obtaining the target user ID is consistent with the method for obtaining the initial user ID.

[0119] Furthermore, the target user is a user whose target portrait is to be obtained.

[0120] Specifically, the target feature data list includes a plurality of target feature data, wherein the target feature data is data related to the target feature that the target user corresponding to the target user ID is involved in before the current moment.

[0121] Furthermore, the target feature is an initial feature screened from a number of initial features.

[0122] Furthermore, the number of features included in the target features is smaller than the number of features included in the initial features.

[0123] S400, input the target feature data list into the target training model to obtain the target portrait corresponding to the target user ID.

[0124] Specifically, the target portrait is all data related to the initial features of the target user before the current moment obtained based on the target training model. It can be understood as: training the model through a large amount of data so that the model can obtain information related to the target user in certain contexts, such as what apps are installed, when they were installed and uninstalled, or what geographical locations they appeared in.

[0125] From the above content, it can be seen that the first embodiment of the present invention can obtain an initial prediction model; obtain a specified user ID list and a specified feature vector list set corresponding to the specified user ID list; input the specified feature vector list set into the initial prediction model to obtain a target priority; continuously adjust the parameters corresponding to the initial prediction model until the target priority meets the preset priority threshold to obtain a target user portrait prediction model; it can be seen that the use of a large language model to process the feature information corresponding to the user solves the problem of large and complex data volume and improves the efficiency of predicting user portraits. When training the large language model, the model prediction value is not directly compared with the true value. The filtered feature information is compared through feature filtering, which improves the accuracy of obtaining the target user portrait prediction model.

[0126] Embodiment 2

[0127] like Figure 5 As shown, this embodiment provides a device for obtaining a user portrait prediction model, the device comprising:

[0128] The first processing module 1 obtains an initial prediction model.

[0129] Specifically, the initial prediction model is a prediction model obtained by inputting the initial user ID list and the target feature vector list set into the initial LLM model for training.

[0130] Furthermore, those skilled in the art know that the methods for training LLM models in the prior art all fall within the protection scope of the present invention and will not be described in detail herein.

[0131] Specifically, the initial user ID list includes a plurality of initial user IDs, wherein the initial user ID is a unique identifier representing an initial user.

[0132] Furthermore, the initial user is a user of a device with a communication function used for model training.

[0133] Furthermore, those skilled in the art know that any method for obtaining a user identifier in the prior art falls within the protection scope of the present invention and will not be described in detail herein.

[0134] Specifically, the target feature vector list set includes several target feature vector lists, wherein each initial user ID corresponds to one target feature vector list, and the target feature vector list includes several target feature vectors.

[0135] Specifically, the first processing module 1 further includes:

[0136] The first acquisition module is used to acquire an initial feature data list corresponding to each initial user ID.

[0137] Specifically, the initial feature data list includes a plurality of initial feature data, wherein the initial feature data is data related to the initial feature that the initial user corresponding to the initial user ID is involved in before the current moment.

[0138] Furthermore, the initial features are the types of data related to the initial user obtained through the device of the initial user, such as: a list of installed apps, the names of installed apps and the time of installation, the names of uninstalled apps and the time of uninstallation, a list of used apps, geographic location information that has appeared, descriptions corresponding to installed apps, descriptions corresponding to uninstalled apps, and other initial features.

[0139] The second acquisition module is used to acquire the first data list and the second data list corresponding to the initial user ID according to the initial feature data list.

[0140] Specifically, the first data list includes a plurality of first data, wherein the first data is sequence data and the like.

[0141] Further, when the initial feature data in the initial feature data list can be converted into a sequence, the sequence corresponding to the initial feature data obtained is the first data, for example: the name of the installed APP and the time of installation.

[0142] Specifically, the second data list includes a plurality of second data, wherein the second data is text data.

[0143] Further, when the initial feature data in the initial feature data list is text, the acquired text corresponding to the initial feature data is the second data, for example: a description of an APP, a message pushed to a user device, and the like.

[0144] The third acquisition module is used to acquire a target feature vector list corresponding to each initial user ID according to the first data list and the second data list.

[0145] Specifically, the target feature vector list includes several target feature vectors.

[0146] Specifically, the third acquisition module further includes:

[0147] The first target feature vector acquisition module is used to acquire the first target feature vector according to the first data list.

[0148] Specifically, the first target feature vector is a vector obtained by converting the first data in the first data list into a vector and then adding them together. It can be understood as: obtaining the i-th first data vector A i =(A i1 , ..., A ij , ..., A im ), A ij is the jth bit in the i-th first data vector, the value range of j is 1 to m, m is the dimension of the first data vector, and the first target feature vector B=(B1, ..., B j , ..., B m ), where B j =∑ n i=1 A ij , the value range of i is 1 to n, and n is the number of first data in the first data list.

[0149] Furthermore, those skilled in the art know that any method for converting a sequence into a vector in the prior art falls within the protection scope of the present invention and will not be described in detail herein.

[0150] The second target feature vector acquisition module is used to acquire the second target feature vector according to the second data list.

[0151] Specifically, the second target feature vector is a vector obtained by converting the second data in the second data list into vectors and then adding them.

[0152] Furthermore, the second target feature vector is obtained in the same manner as the first target feature vector.

[0153] The target feature vector acquisition module is used to acquire the target feature vector according to the first target feature vector and the second target feature vector.

[0154] Specifically, the target feature vector is a vector obtained by adding the first target feature vector and the second target feature vector.

[0155] Furthermore, the method for acquiring the target feature vector is consistent with the method for acquiring the first target feature vector.

[0156] From the above content, we can see that the feature data of different types corresponding to users are uniformly converted into vectors for the training of the large language model, which avoids the problem of non-standardized data caused by diverse sources and formats of user data. The corresponding semantic relationships are understood by the large language model, the correlation between data is obtained, and the accuracy of obtaining the target user portrait is improved.

[0157] The second processing module 2 is used to obtain a specified user ID list and a specified feature vector list set corresponding to the specified user ID list.

[0158] Specifically, the designated user ID list includes a plurality of designated user IDs, wherein the designated user ID is a unique identifier representing a designated user.

[0159] Furthermore, the designated user is a user having a communication device for detecting the initial prediction model.

[0160] Furthermore, the method for obtaining the designated user ID is consistent with the method for obtaining the initial user ID.

[0161] Specifically, the designated feature vector list set includes a plurality of designated feature vector lists, wherein each designated user ID included in the designated user ID list corresponds to a designated feature vector list, and the designated feature vector list includes a plurality of designated feature vectors.

[0162] Specifically, Figure 6 As shown, the second processing module 2 also includes:

[0163] The first designated data list acquisition module 21 is used to acquire the first designated data list corresponding to each designated user ID.

[0164] Specifically, the first designated data list includes a plurality of first designated data, wherein the first designated data is data related to the initial features and involved in a designated user corresponding to a designated user ID within a preset time period.

[0165] Furthermore, the end time of the preset time period is before the current time.

[0166] Furthermore, those skilled in the art know that the time span corresponding to the preset time period can be selected according to actual needs, which falls within the protection scope of the present invention and will not be described in detail here.

[0167] The first designated feature vector acquisition module 22 is used to acquire the first designated feature vector according to the first designated data list.

[0168] Specifically, the method for acquiring the first designated feature vector is consistent with the method for acquiring the target feature vector.

[0169] The second designated data list acquisition module 23 is used to acquire the second designated data list corresponding to each designated user ID.

[0170] Specifically, the second designated data list includes a plurality of second designated data, wherein the second designated data is data related to the initial features and involved in a designated user corresponding to a designated user ID within a designated time period.

[0171] Furthermore, the start time of the specified time period is the end time of the preset time period.

[0172] Furthermore, the end time of the specified time period is the current time.

[0173] The second designated feature vector acquisition module 24 is used to acquire the second designated feature vector according to the second designated data list.

[0174] Specifically, the method for acquiring the second designated feature vector is consistent with the method for acquiring the target feature vector.

[0175] The designated feature vector acquisition module 25 is used to acquire the designated feature vector according to the first designated feature vector and the second designated feature vector.

[0176] Specifically, the designated feature vector includes a first designated feature vector and a second designated feature vector.

[0177] The third processing module 3 is used to input the specified feature vector list set into the initial prediction model to obtain the target priority.

[0178] Specifically, Figure 7 As shown, the third processing module 3 also includes:

[0179] The first intermediate feature vector list acquisition module 31 is used to acquire the first intermediate feature vector list corresponding to the specified user ID list.

[0180] Specifically, the first intermediate feature vector list includes a plurality of first intermediate feature vectors, wherein each designated user ID corresponds to a first intermediate feature vector.

[0181] Furthermore, the first intermediate feature vector is a feature vector corresponding to a specified time period obtained by inputting the first specified feature vector into an initial prediction model for prediction.

[0182] The second intermediate feature vector list acquisition module 32 is used to acquire the second intermediate feature vector list corresponding to the specified user ID list.

[0183] Specifically, the second intermediate feature vector list includes a plurality of second intermediate feature vectors, wherein each designated user ID corresponds to a second intermediate feature vector.

[0184] Furthermore, the second intermediate feature vector is a second designated feature vector corresponding to the acquired designated user ID.

[0185] The target acquisition module 33 is used to acquire the target priority according to the first intermediate feature vector list and the second intermediate feature vector list.

[0186] Specifically, Figure 8 As shown, the target acquisition module 33 also includes:

[0187] The first key feature vector list acquisition module 331 is used to acquire the first key feature vector list according to the first intermediate feature vector list, that is, to acquire the first key feature vector list H={H1, ..., H g , ..., H h}, H g is the gth first key eigenvector, the value range of g is 1 to h, and h is the number of first key eigenvectors.

[0188] Specifically, the first key feature vector list includes a first key feature vector, wherein the first key feature vector is a feature vector corresponding to a specific feature obtained from the first intermediate feature vector.

[0189] Furthermore, the specific feature is an initial feature in which an initial feature corresponding to the first intermediate feature vector overlaps with an initial feature corresponding to the second intermediate feature vector.

[0190] The second key feature vector list acquisition module 332 is used to acquire the second key feature vector list according to the second intermediate feature vector list, that is, to acquire the second key feature vector list P={P1, ..., P g , ..., P h}, P g is the g-th second key eigenvector.

[0191] Specifically, the second key feature vector list includes a second key feature vector, wherein the second key feature vector is a feature vector corresponding to a specific feature obtained from the second intermediate feature vector.

[0192] The target priority acquisition module 333 is used to acquire the target priority according to the first key feature vector list and the second key feature vector list, that is, to acquire the target priority F according to H and P, wherein F meets the following conditions:

[0193] F=∑ h g=1 Q g , Q g H g With P g The similarity between .

[0194] Specifically, those skilled in the art know that any method for calculating the similarity between vectors in the prior art falls within the protection scope of the present invention and will not be described in detail herein.

[0195] The fourth processing module 4 is used to continuously adjust the parameters corresponding to the initial prediction model until the target priority meets a preset priority threshold to obtain a target user portrait prediction model.

[0196] Specifically, the preset priority threshold value ranges from 0.75 to 0.9, wherein those skilled in the art know that the preset priority threshold value can be selected according to actual needs, all of which fall within the protection scope of the present invention and will not be described in detail here.

[0197] From the above content, we can see that the use of a large language model to process the feature information corresponding to the user solves the problem of large and complex data volume and improves the efficiency of predicting user portraits. When training the large language model, the model prediction value is not directly compared with the true value. The feature information after feature screening is compared through feature screening, which improves the accuracy of the target user portrait prediction model.

[0198] Specifically, the device further includes:

[0199] The second execution module is used to build a target training model according to the target feature vector list set.

[0200] Specifically, the second execution module further includes:

[0201] The first candidate module is used to input the target feature vector list set into the initial LLM model for training.

[0202] The second candidate module is used to continuously adjust the parameters of the initial LLM model and screen the initial features to adjust the target feature vector list set until the error is less than a preset threshold to obtain the target training model.

[0203] Specifically, the preset threshold value ranges from 0.2 to 0.3, wherein those skilled in the art know that the preset threshold value can be selected according to actual needs, all of which fall within the protection scope of the present invention and will not be described in detail here.

[0204] Specifically, those skilled in the art know that any method of adjusting model parameters and screening features to change model input in the prior art falls within the protection scope of the present invention and will not be described in detail herein.

[0205] The third execution module is used to obtain a target feature data list corresponding to the target user ID.

[0206] Specifically, the target user ID is a unique identifier that represents the target user, wherein the method for obtaining the target user ID is consistent with the method for obtaining the initial user ID.

[0207] Furthermore, the target user is a user whose target portrait is to be obtained.

[0208] Specifically, the target feature data list includes a plurality of target feature data, wherein the target feature data is data related to the target feature that the target user corresponding to the target user ID is involved in before the current moment.

[0209] Furthermore, the target feature is an initial feature screened from a number of initial features.

[0210] Furthermore, the number of features included in the target features is smaller than the number of features included in the initial features.

[0211] The fourth execution module is used to input the target feature data list into the target training model to obtain the target portrait corresponding to the target user ID.

[0212] Specifically, the target portrait is all data related to the initial features of the target user before the current moment obtained based on the target training model. It can be understood as: training the model through a large amount of data so that the model can obtain information related to the target user in certain contexts, such as what apps are installed, when they were installed and uninstalled, or what geographical locations they appeared in.

[0213] Embodiment 3

[0214] An embodiment of the present invention provides a non-transitory computer-readable storage medium, wherein at least one instruction or at least one program is stored in the non-transitory computer-readable storage medium, and the at least one instruction or at least one program is loaded and executed by a processor to implement the steps:

[0215] Get the initial prediction model.

[0216] Obtaining a designated user ID list and a designated feature vector list set corresponding to the designated user ID list, wherein the designated user ID list includes a plurality of designated user IDs, and obtaining the designated user ID list and a designated feature vector list set corresponding to the designated user ID list includes the following steps:

[0217] A first designated data list corresponding to each designated user ID is obtained, wherein the first designated data list includes a plurality of first designated data, wherein the first designated data is data related to initial features involved in a designated user corresponding to the designated user ID within a preset time period.

[0218] According to the first specified data list, a first specified feature vector is obtained.

[0219] A second designated data list corresponding to each designated user ID is obtained, wherein the second designated data list includes a plurality of second designated data, wherein the second designated data is data related to the initial features involved by the designated user corresponding to the designated user ID within a designated time period.

[0220] According to the second specified data list, a second specified feature vector is obtained.

[0221] A designated feature vector is obtained according to the first designated feature vector and the second designated feature vector.

[0222] Input the specified feature vector list set into the initial prediction model to obtain the target priority.

[0223] The parameters corresponding to the initial prediction model are continuously adjusted until the target priority meets the preset priority threshold to obtain the target user portrait prediction model.

[0224] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0225] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0226] Embodiment 4

[0227] An embodiment of the present invention provides an electronic device, which includes a processor and the non-transitory computer-readable storage medium in Embodiment 3 of the present invention.

[0228] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for obtaining a user portrait prediction model, characterized in that: The method comprises the following steps: Obtaining an initial prediction model; Obtaining a designated user ID list and a designated feature vector list set corresponding to the designated user ID list, wherein the designated user ID list includes a plurality of designated user IDs, and obtaining the designated user ID list and a designated feature vector list set corresponding to the designated user ID list includes the following steps: Obtain a first designated data list corresponding to each designated user ID, wherein the first designated data list includes a plurality of first designated data, wherein the first designated data is data related to initial features involved by the designated user corresponding to the designated user ID within a preset time period, and the initial features include an installed APP list; According to the first specified data list, obtaining a first specified feature vector; Obtain a second designated data list corresponding to each designated user ID, wherein the second designated data list includes a plurality of second designated data, wherein the second designated data is data related to the initial feature and involved in the designated user corresponding to the designated user ID within a designated time period; According to the second specified data list, obtaining a second specified feature vector; Obtaining a designated feature vector according to the first designated feature vector and the second designated feature vector; Inputting the specified feature vector list set into the initial prediction model to obtain the target priority, wherein inputting the specified feature vector list set into the initial prediction model to obtain the target priority includes the following steps: Obtaining a first intermediate feature vector list corresponding to a specified user ID list, wherein the first intermediate feature vector list includes a plurality of first intermediate feature vectors, and the first intermediate feature vector is a feature vector corresponding to a specified time period obtained by inputting the first specified feature vector into an initial prediction model for prediction; Obtain a second intermediate feature vector list corresponding to the specified user ID list, wherein the second intermediate feature vector list includes a plurality of second intermediate feature vectors, and the second intermediate feature vector is a second specified feature vector corresponding to the obtained specified user ID; According to the first intermediate feature vector list and the second intermediate feature vector list, the target priority is obtained, wherein according to the first intermediate feature vector list and the second intermediate feature vector list, the target priority is obtained including the following steps: According to the first intermediate feature vector list, a first key feature vector list is obtained, that is, according to the first intermediate feature vector list, a first key feature vector list H={H1, ..., H g , ..., H h }, H g is the gth first key eigenvector, the value range of g is 1 to h, and h is the number of first key eigenvectors; According to the second intermediate feature vector list, a second key feature vector list is obtained, that is, according to the second intermediate feature vector list, a second key feature vector list P={P1, ..., P g , ..., P h }, P g is the g-th second key eigenvector; According to the first key feature vector list and the second key feature vector list, the target priority is obtained, that is, according to H and P, the target priority F is obtained, where F meets the following conditions: F=∑ h g=1 Q g , Q g H g With P g The similarity between The parameters corresponding to the initial prediction model are continuously adjusted until the target priority meets the preset priority threshold to obtain the target user portrait prediction model.

2. The method for obtaining a user portrait prediction model according to claim 1, characterized in that: The designated feature vector list set includes a plurality of designated feature vector lists, wherein each designated user ID included in the designated user ID list corresponds to a designated feature vector list, and the designated feature vector list includes a plurality of designated feature vectors.

3. The method for obtaining a user portrait prediction model according to claim 1, characterized in that: The first designated data list includes a plurality of first designated data, wherein the first designated data is data related to initial features of a designated user corresponding to an acquired designated user ID within a preset time period.

4. The method for obtaining a user portrait prediction model according to claim 1, characterized in that: The second designated data list includes a plurality of second designated data, wherein the second designated data is data related to the initial features within a designated time period of a designated user corresponding to the acquired designated user ID.

5. A device for obtaining a user portrait prediction model, characterized in that: The device comprises: A first processing module is used to obtain an initial prediction model; A second processing module is used to obtain a specified user ID list and a specified feature vector list set corresponding to the specified user ID list, wherein the second processing module also includes: A first designated data list acquisition module, used to acquire a first designated data list corresponding to each designated user ID, wherein the first designated data list includes a plurality of first designated data, wherein the first designated data is data related to initial features involved by the designated user corresponding to the designated user ID within a preset time period, and the initial features include an installed APP list; A first designated feature vector acquisition module, used for acquiring a first designated feature vector according to a first designated data list; A second designated data list acquisition module is used to acquire a second designated data list corresponding to each designated user ID, wherein the second designated data list includes a plurality of second designated data, wherein the second designated data is data related to the initial features involved by the designated user corresponding to the designated user ID within a designated time period; A second designated feature vector acquisition module, used for acquiring a second designated feature vector according to a second designated data list; A designated feature vector acquisition module, used for acquiring a designated feature vector according to the first designated feature vector and the second designated feature vector; The third processing module is used to input the specified feature vector list set into the initial prediction model to obtain the target priority, wherein the third processing module also includes: A first intermediate feature vector list acquisition module is used to acquire a first intermediate feature vector list corresponding to a specified user ID list, wherein the first intermediate feature vector list includes a plurality of first intermediate feature vectors, and the first intermediate feature vector is a feature vector corresponding to a specified time period obtained by inputting the first specified feature vector into an initial prediction model for prediction; A second intermediate feature vector list acquisition module, used to acquire a second intermediate feature vector list corresponding to a specified user ID list, wherein the second intermediate feature vector list includes a plurality of second intermediate feature vectors, and the second intermediate feature vector is a second specified feature vector corresponding to the acquired specified user ID; A target acquisition module is used to acquire a target priority according to the first intermediate feature vector list and the second intermediate feature vector list, wherein the target acquisition module further includes: The first key feature vector list acquisition module is used to acquire the first key feature vector list according to the first intermediate feature vector list, that is, to acquire the first key feature vector list H={H1, ..., H g , ..., H h }, H g is the gth first key eigenvector, the value range of g is 1 to h, and h is the number of first key eigenvectors; The second key feature vector list acquisition module is used to acquire the second key feature vector list according to the second intermediate feature vector list, that is, to acquire the second key feature vector list P={P1, ..., P g , ..., P h }, P g is the g-th second key eigenvector; The target priority acquisition module is used to acquire the target priority according to the first key feature vector list and the second key feature vector list, that is, to acquire the target priority F according to H and P, wherein F meets the following conditions: F=∑ h g=1 Q g , Q g H g With P g The similarity between The fourth processing module is used to continuously adjust the parameters corresponding to the initial prediction model until the target priority meets the preset priority threshold to obtain the target user portrait prediction model.

6. The device for obtaining a user portrait prediction model according to claim 5, characterized in that: The designated feature vector list set includes a plurality of designated feature vector lists, wherein each designated user ID included in the designated user ID list corresponds to a designated feature vector list, and the designated feature vector list includes a plurality of designated feature vectors.

7. The device for obtaining a user portrait prediction model according to claim 5, characterized in that: The first designated data list includes a plurality of first designated data, wherein the first designated data is data related to initial features of a designated user corresponding to an acquired designated user ID within a preset time period.

8. The device for obtaining a user portrait prediction model according to claim 5, characterized in that: The second designated data list includes a plurality of second designated data, wherein the second designated data is data related to the initial features within a designated time period of a designated user corresponding to the acquired designated user ID.

9. A non-transitory computer-readable storage medium, wherein at least one instruction or at least one program is stored in the non-transitory computer-readable storage medium, characterized in that: The at least one instruction or the at least one program is loaded and executed by the processor to implement the method for obtaining a user portrait prediction model as described in any one of claims 1-4.

10. An electronic device, characterized in that: The invention comprises a processor and the non-transitory computer-readable storage medium as claimed in claim 9.

Images