Operation prediction method and device, electronic equipment and storage medium

Abstract

The present disclosure relates to a kind of operation prediction method, device, electronic equipment and storage medium, the method can be applied to artificial intelligence technical field, the method includes: obtaining the first operation characteristic sequence corresponding to target object;First operation characteristic sequence is split and handled, and first target feature sequence and second target feature sequence are obtained;First target feature sequence is input to the encoder in operation prediction model and is encoded, and third target feature sequence is obtained;Second target feature sequence and third target feature sequence are input to the decoder in operation prediction model and are decoded, and the target operation information corresponding to target object in the time to be predicted is obtained.Utilize the encoder and decoder in the embodiment of the present disclosure by adopting, can make operation prediction model better adapt to the first operation characteristic sequence of long time span, can improve the accuracy of target operation information prediction for target object in the time to be predicted.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to an operation prediction method, apparatus, electronic device, and storage medium. Background Technology

[0002] In most industries, especially the internet industry, massive amounts of data are generated daily. Users generate a large amount of log information every moment; within applications, users perform different actions at different times each day, which are written to the background system logs. We call this data, received by the system at different times and describing the changes in one or more user characteristics over time, time-series data. We can predict future user actions using past time-series data. Related technologies can use trained neural networks to predict future user actions; however, these technologies struggle to learn user action characteristics effectively for long-term time-series data, resulting in low accuracy in action prediction. Summary of the Invention

[0003] In view of the aforementioned technical problems, this disclosure proposes an operation prediction method, apparatus, electronic device, and storage medium.

[0004] According to one aspect of the embodiments of the present disclosure, an operation prediction method is provided, the method comprising: Obtain a first operation feature sequence corresponding to the target object, wherein the first operation feature sequence characterizes the operation features of the target object in a first preset time period; The first operational feature sequence is segmented to obtain a first target feature sequence and a second target feature sequence; The first target feature sequence is input into the encoder in the operational prediction model for encoding processing to obtain the third target feature sequence; The second target feature sequence and the third target feature sequence are input into the decoder in the operation prediction model for operation decoding processing to obtain the target operation information corresponding to the target object at the time to be predicted.

[0005] According to another aspect of the embodiments of this disclosure, an operation prediction apparatus is provided, comprising: The sequence acquisition module is used to acquire a first operation feature sequence corresponding to the target object, wherein the first operation feature sequence characterizes the operation features of the target object in a first preset time period. The preprocessing module is used to segment the first operational feature sequence to obtain a first target feature sequence and a second target feature sequence. The execution module is used to input the first target feature sequence into the encoder in the operation prediction model for encoding processing to obtain the third target feature sequence; The operation prediction module is used to input the second target feature sequence and the third target feature sequence into the decoder in the operation prediction model for operation decoding processing, so as to obtain the target operation information corresponding to the target object at the time to be predicted.

[0006] According to another aspect of the present disclosure, an electronic device is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the above-described operation prediction method.

[0007] According to another aspect of the present disclosure, a computer-readable storage medium is provided that, when instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to perform the above-described operation prediction method.

[0008] According to another aspect of the present disclosure, a computer program product containing instructions is provided that, when run on a computer, causes the computer to perform the above-described operation prediction method.

[0009] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects: By segmenting the first operational feature sequence of the target object, the segmented first target feature sequence is input into the encoder in the operation prediction model for encoding processing to obtain the third target feature sequence. The second and third target feature sequences are then input into the decoder in the operation prediction model for operation decoding processing to obtain the target operation information of the target object at the time to be predicted. By using an encoder and decoder, the operation prediction model can be better applied to the first operational feature sequence with a long time span, which can improve the accuracy of predicting the target operation information of the target object at the time to be predicted, thereby providing services better according to user needs and improving user experience.

[0010] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0011] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0012] Figure 1 This is a schematic diagram illustrating an application environment according to an exemplary embodiment; Figure 2 This is a schematic diagram illustrating an operational prediction model according to an exemplary embodiment; Figure 3This is a flowchart illustrating an operation prediction method according to an exemplary embodiment; Figure 4 This is a flowchart illustrating a method for inputting a first target feature sequence into an encoder in an operational prediction model for encoding processing to obtain a third target feature sequence, according to an exemplary embodiment. Figure 5 This is a flowchart of a method according to an exemplary embodiment, which inputs a second target feature sequence and a third target feature sequence into a decoder in an operation prediction model for operation decoding processing to obtain target operation information corresponding to the target object at the time to be predicted. Figure 6 This is a flowchart illustrating a method for obtaining a first operational feature sequence corresponding to a target object according to an exemplary embodiment; Figure 7 This is a flowchart illustrating a method for training a preset prediction model based on a second operation feature sequence to obtain a trained operation prediction model, according to an exemplary embodiment. Figure 8 This is a flowchart illustrating a method for determining at least one set of sample feature sequences and operation information labels corresponding to at least one set of sample feature sequences based on a second operation feature sequence, according to an exemplary embodiment. Figure 9 This is a flowchart illustrating a method for inputting a first sample feature sequence into a preset encoder for encoding processing to obtain a third sample feature sequence, according to an exemplary embodiment. Figure 10 This is a structural block diagram illustrating an operation prediction device according to an exemplary embodiment; Figure 11 This is a block diagram illustrating an electronic device for predicting target operation information of a target object according to an exemplary embodiment; Figure 12 This is a block diagram illustrating an electronic device for predicting target operation information of a target object according to an exemplary embodiment. Detailed Implementation

[0013] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0014] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0015] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or computers-controlled machines to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. AI software technology mainly includes computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0016] In recent years, with the research and progress of artificial intelligence technology, it has been widely applied in many fields. The solutions provided in this application involve technologies such as machine learning / deep learning, which are specifically illustrated in the following embodiments: Please refer to Figure 1 , Figure 1 This diagram illustrates an application system according to an embodiment of this application. The application system can be used in the operation prediction method of this application. Figure 1 As shown, the application system may include at least server 01 and terminal 02.

[0017] In this embodiment, server 01 can be used to train a preset prediction model. Specifically, server 01 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms.

[0018] In this embodiment, terminal 02 can predict operation information based on the operation prediction model trained by server 01. Terminal 02 may include physical devices such as smartphones, desktop computers, tablets, laptops, smart speakers, in-vehicle terminals, digital assistants, augmented reality (AR) / virtual reality (VR) devices, and smart wearable devices, and may also include software running on the physical device, such as applications. The operating system running on terminal 02 in this embodiment may include, but is not limited to, Android, iOS, Linux, and Windows. This invention can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, and assisted driving.

[0019] In addition, it should be noted that, Figure 1 The example shown is merely one application environment provided by this disclosure. In practical applications, other application environments may also be included. For example, the training of a preset prediction model can also be implemented on terminal 02.

[0020] In the embodiments described in this specification, the terminal 02 and the server 01 can be directly or indirectly connected through wired or wireless communication, and this application does not limit this connection.

[0021] It should be noted that the following diagram shows one possible sequence of steps, and it is not strictly necessary to follow this order. Some steps can be executed in parallel without interdependence.

[0022] See Figure 2 , Figure 2 This is a schematic diagram illustrating an operational prediction model according to an exemplary embodiment. For example... Figure 2 As shown, the operation prediction model may include an encoder and a decoder; the encoder may include a first input layer and an encoding layer; the decoder may include a second input layer, a first decoding layer, and a second decoding layer. The operation prediction model may be obtained by training a preset prediction model. The preset prediction model may include a preset encoding layer and a preset decoding layer; the preset encoding layer may include a first input layer and an encoding layer; the preset decoder may include a second input layer, a first decoding layer, and a second decoding layer.

[0023] See Figure 3 , Figure 3 This is a flowchart illustrating an operation prediction method according to an exemplary embodiment, such as... Figure 3 As shown, the operation prediction method may include the following steps.

[0024] In step S301, the first operation feature sequence corresponding to the target object is obtained.

[0025] In one specific embodiment, the target object may be a user account. The first operation feature sequence can characterize the operation features of the target object within a first preset time period. Specifically, the first operation feature sequence can be a sequence of operation features within the first preset time period ordered chronologically. Different operation types can correspond to different operation features. For example, operation types may include keyword search operations, category filtering operations, or consumption operations; the operation feature corresponding to a keyword search operation can be the feature corresponding to the searched keyword information; the operation feature corresponding to a category filtering operation can be the feature corresponding to the filtered category information; and the operation feature corresponding to a consumption operation can be the feature corresponding to the consumption information.

[0026] In one specific embodiment, the operation records of the target object within a historical time period can be obtained. Then, based on the operation records of the target object within the historical time period, the historical operation feature sequence corresponding to the target object can be determined through feature extraction processing. Optionally, the above-mentioned historical operation feature sequence can be used as the first operation feature sequence corresponding to the target object. Correspondingly, the first preset time period and the historical time period are the same time period. Optionally, the operation feature sequence corresponding to the first preset time period in the above-mentioned historical operation feature sequence can be used as the first operation feature sequence. The first preset time period is the time period corresponding to the current time and the preset time in the above-mentioned historical time period, and the preset time is later than the earliest time in the above-mentioned historical time period.

[0027] In step S302, the first operational feature sequence is segmented to obtain the first target feature sequence and the second target feature sequence.

[0028] In one specific embodiment, the time range corresponding to the first operational feature sequence may include the time range corresponding to the first target feature sequence and the time range corresponding to the second target feature sequence.

[0029] In a specific embodiment, the segmentation position information corresponding to the first operation feature sequence can be determined based on the time span information of the first operation feature sequence. The first operation feature sequence can be segmented based on the segmentation position information to obtain the first target feature sequence and the second target feature sequence.

[0030] The time span information of the first operational feature sequence can characterize the sequence length of the first operational feature sequence. The time span information of the first operational feature sequence can refer to the difference between the earliest and latest operation times in the sequence. The segmentation position information corresponding to the first operational feature sequence can be one of multiple operation times corresponding to the first operational feature sequence. Specifically, the segmentation position information corresponding to the first operational feature sequence can be determined based on the time span information and a preset ratio. For example, the time span information of the first operational feature sequence can be multiplied by the preset ratio, and the result can be added to the earliest operation time in the first operational feature sequence to obtain the segmentation position information corresponding to the first operational feature sequence. For example, the preset ratio can be 0.5. Based on the above segmentation position information, the operational feature sequences located on both sides of the segmentation position information in the first operational feature sequence are respectively used as the first target feature sequence and the second target feature sequence, wherein the operational feature corresponding to the segmentation position information can belong to either the first target feature sequence or the second target feature sequence.

[0031] In the above embodiments, by controlling the segmentation position information corresponding to the first operation feature sequence through preset ratio information, it can be ensured that both the encoder and decoder in the operation prediction model can obtain sufficient features about the target object, thereby ensuring the prediction accuracy of the target operation information.

[0032] In step S303, the first target feature sequence is input into the encoder in the operational prediction model for encoding processing to obtain the third target feature sequence.

[0033] In a specific embodiment, such as Figure 4 As shown, step S303 above may include: S401. Based on the target delay scalar lookup table in the operation prediction model, find the delay scalar information corresponding to each operation feature in the first target feature sequence, and obtain the first target delay scalar information corresponding to the first target feature sequence.

[0034] In one specific embodiment, the target delay scalar lookup table can represent the mapping relationship between different operation features and target delay scalar information. The target delay scalar lookup table can be a dense vector lookup table. Specifically, each delay scalar information in the target delay scalar lookup table can be a vector of a preset dimension, and the length of the target delay scalar lookup table can be the same as the length of the historical operation feature sequence; the preset dimension can be set according to actual needs, for example, the preset dimension can be 64 or 128. The first target delay scalar information can be a matrix. The first target delay scalar information can represent the operation feature corresponding to the latest operation time in the first target feature sequence.

[0035] For example, suppose the first target feature sequence is x 1, x 2, ..., x t Each operational feature can be found through the target delay scalar lookup table. x i corresponding e i Accordingly, the first target delay scalar information can be obtained as [ e 1, e 2, ..., e t ].

[0036] In the above embodiments, the first target delay scalar information corresponding to the first target feature sequence is obtained based on the target delay scalar lookup table. The operation feature corresponding to the latest operation time in the first target feature sequence is characterized by the first target delay scalar information, which can better learn the operation feature in the first target feature sequence.

[0037] S402. Based on the position information of each operational feature in the first target feature sequence within the first target feature sequence, determine the first target position scalar information corresponding to the first target feature sequence.

[0038] In one specific embodiment, the position information of each operational feature in the first target feature sequence can be the number of times that operational feature appears in the first target feature sequence in chronological order. For example, if the first target feature sequence is arranged in chronological order from earliest to latest, then the position information of each operational feature in the first target feature sequence can be the number of times that operational feature appears in the first target feature sequence in chronological order; correspondingly, if the first target feature sequence is arranged in chronological order from latest to earliest, then the position information of each operational feature in the first target feature sequence can be the number of times that operational feature appears in the first target feature sequence in chronological order from latest to earliest. The first target position scalar information can be used to characterize the position information of the operational features in the first target feature sequence.

[0039] In a specific embodiment, the positional scalar information corresponding to each operational feature in the first target feature sequence can be obtained through the positional information of each operational feature in the first target feature sequence; based on the positional scalar information corresponding to each operational feature in the first target feature sequence, the positional scalar information of the first target can be obtained. Specifically, the positional scalar information corresponding to each operational feature in the first target feature sequence can be calculated using the following formula:

[0040] Where i represents the position information of the operational feature in the first target feature sequence; D represents the preset dimension.

[0041] For example, suppose the first target feature sequence is x 1, x 2, ..., x t Therefore, according to the above formula, the positional scalar information corresponding to the operational features in the first target feature sequence can be obtained. p 1, p 2, ..., p t Accordingly, the scalar information of the first target location can be obtained as [ p 1, p 2, ..., p t ].

[0042] S403. The delay scalar information and the position scalar information of the first target are concatenated to obtain the feature sequence of the fourth target.

[0043] In one specific embodiment, the fourth target feature sequence can be obtained by concatenating the first target delay scalar information and the first target position scalar information. For example, suppose the first target delay scalar information is [ e 1, e 2, e 3], the first target location scalar information is [ p 1, p 2, p [3] By concatenating the first target delay scalar information and the first target position scalar information, the fourth target feature sequence can be obtained. e 1, e 2, e 3, p 1, p 2, p 3).

[0044] In one specific embodiment, the above steps S401-S403 may be performed in the first input layer of the encoder that operates the prediction model.

[0045] S404. Input the fourth target feature sequence into the encoding layer of the encoder for encoding processing to obtain the third target feature sequence.

[0046] In a specific embodiment, based on the aforementioned fourth target feature sequence X, three vector sequences (Q, K, and V) are generated according to the following formula:

[0047]

[0048]

[0049] Where X is the fourth target feature sequence, This is the parameter matrix. It can be understood that the above parameter matrix represents the model parameters in the encoding layer of the encoder that operates the predictive model.

[0050] Based on the above three vector sequences, the third target feature sequence S can be obtained using the following formula:

[0051] Where Q, K, and V are three vector sequences generated based on the fourth target feature sequence X. softmax ( ) represents the normalized exponential function, and D represents the preset dimension.

[0052] In the above embodiments, by encoding the first target feature sequence through the encoder in the operation prediction model, the variable-length first target feature sequence can be transformed into a third target feature sequence of fixed length that encodes the operation features of the target object. This allows the operation prediction model to be better applied to the first operation feature sequence with a long time span, thereby improving the prediction accuracy.

[0053] In step S304, the second target feature sequence and the third target feature sequence are input into the decoder in the operation prediction model for operation decoding processing to obtain the target operation information corresponding to the target object at the time to be predicted.

[0054] In one specific embodiment, the time to be predicted can be the next moment after the latest moment in the first preset time period. Target operation information can refer to the operation information of the target object at the time to be predicted, predicted by the operation prediction model.

[0055] In a specific embodiment, the operation prediction model can obtain the probability of occurrence of different operation information for the target object at the time to be predicted, and select the operation information with the highest probability of occurrence from the above different operation information as the target operation information.

[0056] In a specific embodiment, such as Figure 5 As shown, step S304 above may include: S501. Based on the target delay scalar lookup table in the operation prediction model, find the delay scalar information corresponding to each operation feature in the second target feature sequence, and obtain the second target delay scalar information corresponding to the second target feature sequence.

[0057] In a specific embodiment, the second target delay scalar information can characterize the operation feature corresponding to the latest operation time in the second target feature sequence.

[0058] In one specific embodiment, the target delay scalar lookup table can represent the mapping relationship between different operational features and target delay scalar information. Specifically, the process of obtaining the second target delay scalar information through the target delay scalar lookup table can refer to step S401 above, and will not be repeated here.

[0059] In the above embodiments, the second target delay scalar information corresponding to the second target feature sequence is obtained based on the target delay scalar lookup table. The operation feature corresponding to the latest operation time in the second target feature sequence is characterized by the second target delay scalar information, which can better learn the operation feature in the second target feature sequence.

[0060] S502. Based on the position information of each operational feature in the second target feature sequence within the second target feature sequence, determine the second target position scalar information corresponding to the second target feature sequence.

[0061] In a specific embodiment, the position information of each operational feature in the second target feature sequence can be the number of times that operational feature appears in the second target feature sequence in chronological order. The second target position scalar information can be used to characterize the position information of the operational features in the second target feature sequence. Specifically, the process of determining the second target position scalar information can refer to step S402 described above, and will not be repeated here.

[0062] S503. The delay scalar information of the second target and the position scalar information of the second target are concatenated to obtain the feature sequence of the fifth target.

[0063] In one specific embodiment, the splicing process can refer to step S403 above, and will not be repeated here.

[0064] In one specific embodiment, the above steps S501-S503 may be performed in the second input layer of the decoder that operates the prediction model.

[0065] S504. Input the fifth target feature sequence into the first decoding layer for decoding processing to obtain the sixth target feature sequence.

[0066] In one specific embodiment, based on the fifth target feature sequence X1 described above, the following three vector sequences (Q1, K1, and V1) are generated according to the following formulas:

[0067]

[0068]

[0069] Where X1 is the feature sequence of the fifth target. This is the parameter matrix. It can be understood that the above parameter matrix represents the model parameters of the first decoding layer in the decoder that operates the prediction model.

[0070] Based on the above three vector sequences, the sixth target feature sequence X2 can be obtained using the following formula:

[0071] Wherein, Q1, K1, and V1 are three vector sequences generated based on the fifth target feature sequence X1. softmax ( ) represents the normalized exponential function, and D represents the preset dimension.

[0072] S505. Input the sixth target feature sequence and the third target feature sequence into the second decoding layer for decoding processing to obtain target operation information.

[0073] In a specific embodiment, based on the sixth target feature sequence X2 and the third target feature sequence S, three vector sequences (Q2, K2, and V2, respectively) can be generated:

[0074]

[0075]

[0076] Where X2 is the feature sequence of the sixth target, and S is the feature sequence of the third target. This is the parameter matrix. It can be understood that the above parameter matrix represents the model parameters of the second decoding layer in the decoder that operates the prediction model.

[0077] Based on the three vector sequences Q2, K2, and V2 mentioned above, the target operation information can be obtained using the following formula. :

[0078] Wherein, Q2, K2, and V2 are three vector sequences generated based on the sixth target feature sequence X2 and the third target feature sequence S. softmax ( ) represents the normalized exponential function, and D represents the preset dimension.

[0079] In the above embodiments, by encoding the first target feature sequence through the encoder in the operation prediction model, the variable-length first target feature sequence can be transformed into a third target feature sequence of fixed length that encodes the operation features of the target object. Then, the target operation information is generated by decoding through the decoder, so that the operation prediction model can be better applied to the first operation feature sequence with a long time span, thereby improving the prediction accuracy.

[0080] In the above embodiments, by segmenting the first operation feature sequence of the target object, the segmented first target feature sequence is input into the encoder in the operation prediction model for encoding processing to obtain the third target feature sequence. The second and third target feature sequences are then input into the decoder in the operation prediction model for operation decoding processing to obtain the target operation information of the target object at the time to be predicted. By using an encoder and a decoder, the operation prediction model can be better applied to the first operation feature sequence with a long time span, which can improve the accuracy of predicting the target operation information of the target object at the time to be predicted, thereby providing services better according to user needs and improving user experience.

[0081] In one specific embodiment, the above method may further include: Obtain the preset prediction model.

[0082] In one specific embodiment, the preset prediction model may refer to an untrained operation prediction model.

[0083] In a specific embodiment, such as Figure 6 As shown, prior to step S301, the method may further include: S601. Obtain the historical operation feature sequence corresponding to the target object.

[0084] In one specific embodiment, the historical operation feature sequence can characterize the operation features of the target object within a historical time period. The historical time period may include a first preset time period.

[0085] In one specific embodiment, the operation records of the target object within a historical time period can be obtained, and the historical operation feature sequence corresponding to the target object can be obtained based on the operation records of the target object within the historical time period.

[0086] S602. The historical operation feature sequence is segmented to obtain the first operation feature sequence and the second operation feature sequence.

[0087] In one specific embodiment, the second operational feature sequence can characterize the operational features of the target object within a second preset time period. It is understood that the aforementioned historical time period may include a first preset time period and a second preset time period.

[0088] In a specific embodiment, the segmentation position information corresponding to the historical operation feature sequence can be determined based on the time span information of the historical operation feature sequence. By segmenting the historical operation feature sequence according to the aforementioned segmentation position information, a first operation feature sequence and a second operation feature sequence can be obtained. Specifically, the segmentation position information corresponding to the historical operation feature sequence can be one or two of the multiple operation times corresponding to the historical operation feature sequence. Optionally, for historical operation feature sequences with a long sequence length, the segmentation position information can be two of the multiple operation times. When the aforementioned segmentation position information represents one operation time, the operation feature sequences located on either side of the aforementioned segmentation position information in the historical operation feature sequence are respectively designated as the first operation feature sequence and the second operation feature sequence. The operation feature corresponding to the segmentation position information can belong to either the first or the second operation feature sequence. When the aforementioned segmentation position information represents two operation times, the earliest and latest operation times in the historical operation feature sequence, as well as the two operation times in the segmentation position information, are sorted in chronological order (assuming the sorting result is earliest operation time, first operation time, second operation time, and latest operation time). The operation feature sequence between the earliest and first operation times can be selected as the first operation feature sequence, or the operation feature sequence between the first and second operation times can be selected as the first operation feature sequence, and the operation feature sequence between the second and latest operation times can be selected as the second operation feature sequence. For example, through segmentation processing, the earlier 50% of the operation features in the historical operation feature sequence can be obtained as the second operation feature sequence, and the later 50% of the operation features in the historical operation feature sequence can be obtained as the first operation feature sequence.

[0089] S603. Based on the second operation feature sequence, train the preset prediction model to obtain the trained operation prediction model.

[0090] In a specific embodiment, such as Figure 7 As shown, step S603 above may include: S701. Based on the second operation feature sequence, determine at least one set of sample feature sequences and operation information labels corresponding to at least one set of sample feature sequences.

[0091] In one specific embodiment, each set of sample feature sequences may include a first set of sample feature sequences and a second set of sample feature sequences. Each set of sample feature sequences and its corresponding operation information label can be used to train a preset prediction model.

[0092] In one specific embodiment, the above method may further include: Obtain the time span information of the preset time span upper limit and the second operation feature sequence.

[0093] In one specific embodiment, the preset time span upper limit can refer to the maximum time span of a preset feature sequence. The preset time span upper limit can be set according to actual needs.

[0094] In one specific embodiment, the time span information of the second operation feature sequence can characterize the sequence length of the second operation feature sequence. This time span information can be obtained by taking the difference between the latest operation time and the earliest operation time in the second operation feature sequence.

[0095] In a specific embodiment, such as Figure 8 As shown, step S701 above may include: S801. Based on the time span information of the second operation feature sequence and the preset time span upper limit, the second operation feature sequence is grouped to obtain at least one third operation feature sequence.

[0096] In one specific embodiment, the second operational feature sequence can be grouped based on a preset time span upper limit and the time span information of the second operational feature sequence, such that the time span information of each grouped third operational feature sequence is less than or equal to the preset time span upper limit. Specifically, if the time span information of the second operational feature sequence is greater than the preset time span upper limit, at least two third operational feature sequences can be obtained through grouping; if the time span information of the second operational feature sequence is less than or equal to the preset time span upper limit, the second operational feature sequence can be used as the third operational feature sequence.

[0097] S802. Based on the target operation features in each third operation feature sequence, generate operation information labels corresponding to each third operation feature sequence.

[0098] In one specific embodiment, the target operation feature may be the operation feature with the latest operation time in each third operation feature sequence.

[0099] In a specific embodiment, the target operation feature in each third operation feature sequence can be used as the operation information label corresponding to the third operation feature sequence.

[0100] S803. Extract the operation feature sequence other than the target operation feature from each third operation feature sequence to obtain the fourth operation feature sequence corresponding to each third operation feature sequence.

[0101] In a specific embodiment, the operation feature sequence other than the target operation feature in each third operation feature sequence can be used as the fourth operation feature sequence corresponding to the third operation feature sequence.

[0102] S804. Based on the time span information of each fourth operation feature sequence, determine the segmentation position information corresponding to each fourth operation feature sequence.

[0103] In a specific embodiment, the process of determining the segmentation position information corresponding to each fourth operation feature sequence can refer to step S302 above.

[0104] S805. Based on the segmentation position information corresponding to each fourth operation feature sequence, perform segmentation processing on each fourth operation feature sequence to obtain at least one set of sample feature sequences.

[0105] In a specific embodiment, the segmentation process of each fourth operation feature sequence can refer to step S302 above.

[0106] S702. Input the first sample feature sequence into the preset encoder for encoding processing to obtain the third sample feature sequence.

[0107] In the above embodiments, the second operation feature sequence is grouped based on the time span information of the second operation feature sequence and the preset time span upper limit. This can avoid the second operation feature sequence being too long and affecting the learning of model parameters in the operation prediction model, thereby improving the prediction accuracy of the target operation information of the target object.

[0108] In a specific embodiment, such as Figure 9 As shown, step S702 above may include: S901. Based on the preset delay scalar lookup table in the preset prediction model, find the delay scalar information corresponding to each operation feature in the first sample feature sequence, and obtain the sample delay scalar information corresponding to the first sample feature sequence.

[0109] In one specific embodiment, a preset delay scalar lookup table can represent the mapping relationship between different operation features and preset delay scalar information. Specifically, the preset delay scalar lookup table can be a target delay scalar lookup table that has not been updated; exemplarily, the preset delay scalar lookup table can be obtained after initialization. The preset delay scalar lookup table can be a dense vector lookup table. Each delay scalar information in the preset delay scalar lookup table can be a vector of a preset dimension, and the length of the preset delay scalar lookup table can be the same as the length of the historical operation feature sequence. Sample delay scalar information can be a matrix. Sample delay scalar information can represent the operation feature corresponding to the latest operation time in the first sample feature sequence.

[0110] In the above embodiments, the sample delay scalar information corresponding to the first sample feature sequence is obtained based on the sample delay scalar lookup table. The operation feature corresponding to the latest operation time in the first sample feature sequence is characterized by the sample delay scalar information, which can better learn the operation feature corresponding to the first sample feature sequence.

[0111] In a specific embodiment, the process of obtaining sample delay scalar information through a preset delay scalar lookup table can refer to step S401 above, and will not be repeated here.

[0112] S902. Based on the position information of each operational feature in the first sample feature sequence within the first sample feature sequence, determine the sample position scalar information corresponding to the first sample feature sequence.

[0113] In a specific embodiment, the position information of each operational feature in the first sample feature sequence can be the number of times that operational feature appears in the first sample feature sequence in chronological order. The sample position scalar information corresponding to the first sample feature sequence can characterize the position information of multiple operational features in the first sample feature sequence. Specifically, the process of determining the sample position scalar information can refer to step S402 above, and will not be repeated here.

[0114] S903. The sample delay scalar information and sample position scalar information are concatenated to obtain the fourth sample feature sequence.

[0115] In one specific embodiment, the splicing process can refer to step S403 above, and will not be repeated here.

[0116] In one specific embodiment, the above steps S901-S903 may be completed in the first input layer of the preset encoder.

[0117] S904. Input the fourth sample feature sequence into the encoding layer of the preset encoder for encoding processing to obtain the third sample feature sequence.

[0118] In one specific embodiment, the encoding process of the preset encoder can refer to step S404 above, and will not be repeated here.

[0119] S703. Input the second sample feature sequence and the third sample feature sequence into the preset decoder for operation decoding processing to obtain prediction operation information.

[0120] In a specific embodiment, the predicted operation information may refer to the operation information corresponding to the next operation time in the set of sample feature sequences predicted by the preset decoder through operation decoding processing. Specifically, the operation decoding process can refer to step S304 above, and will not be repeated here.

[0121] In one specific embodiment, a fifth sample feature sequence can be obtained by inputting the second sample feature sequence into the second input layer of the preset decoder; a sixth sample feature sequence can be obtained by inputting the fifth sample feature sequence into the first decoding layer of the preset decoder; and prediction operation information can be obtained by inputting the sixth sample feature sequence and the third sample feature sequence into the second decoding layer of the preset decoder.

[0122] S704. Obtain operation loss information based on predicted operation information and operation information labels.

[0123] In one specific embodiment, operational loss information can characterize the degree of difference between the predicted operational information and the operational information label. Operational loss information can be used to adjust the parameters in a pre-defined prediction model to train the operational prediction model.

[0124] In a specific embodiment, the operation loss information can be obtained using the following formula. :

[0125] in, To predict operational information, For operation information labels, This is a preset threshold. Specifically, It can be set according to actual needs, and no restrictions are imposed here.

[0126] S705. Train a preset prediction model based on the operation loss information to obtain the operation prediction model.

[0127] In one specific embodiment, step S705 above may include: Based on the operational loss information, the model parameters in the preset prediction model and the preset delay scalar information in the preset delay scalar lookup table are updated to obtain the updated preset prediction model. Based on the updated preset prediction model, the above process is repeated: based on the second operation feature sequence, at least one set of sample feature sequences and operation information labels corresponding to at least one set of sample feature sequences are determined and updated based on the operation loss information. The model parameters in the preset prediction model and the preset delay scalar information in the preset delay scalar lookup table are updated to obtain the updated preset prediction model. This process continues until the preset convergence condition is reached. The preset prediction model obtained when the preset convergence condition is reached is then used as the operation prediction model.

[0128] In one specific embodiment, the model parameters in the aforementioned preset prediction model may include the aforementioned multiple parameter matrices, specifically including the parameter matrix in the encoding layer of the preset encoder. The parameter matrix in the first decoding layer of the preset decoder The parameter matrix in the second decoding layer of the preset decoder It is understandable that the target delay scalar lookup table in the operational prediction model is obtained by updating the delay scalar information in the preset delay scalar lookup table.

[0129] In one specific embodiment, the aforementioned preset convergence condition may include the operational loss information being less than a first preset loss threshold, or the loss difference information being less than a second preset loss threshold, etc. The aforementioned loss difference information may be the difference between the operational loss information of two adjacent training iterations.

[0130] In the above embodiments, by acquiring the historical operation feature sequence of the target object, segmenting the historical operation feature sequence to obtain the first operation feature sequence and the second operation feature sequence, and training the preset prediction model based on the second operation feature sequence to obtain the operation prediction model corresponding to the target object, the first operation feature sequence can be input into the operation prediction model to obtain the target operation information of the target object at the time to be predicted. This can avoid the reduction in prediction accuracy due to the excessive length of the historical operation feature sequence, improve the accuracy of predicting the target operation information of the target object at the time to be predicted, and thus better provide services according to user needs and improve user experience.

[0131] Figure 10 This is a structural block diagram illustrating an operation prediction device according to an exemplary embodiment. (Refer to...) Figure 10 The device includes: The sequence acquisition module 1010 is used to acquire the first operation feature sequence corresponding to the target object. The first operation feature sequence represents the operation features of the target object in a first preset time period. The preprocessing module 1020 is used to segment the first operational feature sequence to obtain a first target feature sequence and a second target feature sequence; Execution module 1030 is used to input the first target feature sequence into the encoder in the operation prediction model for encoding processing to obtain the third target feature sequence; The operation prediction module 1040 is used to input the second target feature sequence and the third target feature sequence into the decoder in the operation prediction model for operation decoding processing, so as to obtain the target operation information corresponding to the target object at the time to be predicted.

[0132] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0133] Figure 11This is a block diagram illustrating an electronic device for predicting target operation information of a target object according to an exemplary embodiment. The electronic device may be a server, and its internal structure diagram may be as follows: Figure 11 As shown, the electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an operation prediction method.

[0134] Figure 12 This is a block diagram illustrating an electronic device for predicting target operation information of a target object according to an exemplary embodiment. The electronic device may be a terminal, and its internal structure diagram may be as follows: Figure 12 As shown, the electronic device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an operation prediction method. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.

[0135] Those skilled in the art will understand that Figure 11 or Figure 12 The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the electronic device to which the present disclosure is applied. A specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements. In an exemplary embodiment, an electronic device is also provided, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the operation prediction method as described in the embodiments of this disclosure.

[0136] In an exemplary embodiment, a computer-readable storage medium is also provided, wherein when the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the operation prediction method of the embodiments of the present disclosure.

[0137] In an exemplary embodiment, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform the operation prediction method of the present disclosure embodiments.

[0138] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0139] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0140] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. An operational prediction method, characterized in that, The method includes: Obtain a first operation feature sequence corresponding to the target object. The first operation feature sequence represents the operation features of the target object in a first preset time period. The target object is a user account. Different operation types correspond to different operation features. The operation types include keyword search operation, category filtering operation, or consumption operation. The operation feature corresponding to the keyword search operation is the feature corresponding to the searched keyword information. The operation feature corresponding to the category filtering operation is the feature corresponding to the filtered category information. The operation feature corresponding to the consumption operation is the feature corresponding to the consumption information. The first operational feature sequence is segmented to obtain a first target feature sequence and a second target feature sequence; The first target feature sequence is input into the encoder in the operational prediction model for encoding processing to obtain the third target feature sequence; The second target feature sequence and the third target feature sequence are input into the decoder in the operation prediction model for operation decoding processing to obtain the target operation information corresponding to the target object at the time to be predicted. The step of inputting the first target feature sequence into the encoder in the operational prediction model for encoding processing to obtain the third target feature sequence includes: Based on the target delay scalar lookup table in the operation prediction model, the delay scalar information corresponding to each operation feature in the first target feature sequence is looked up to obtain the first target delay scalar information corresponding to the first target feature sequence. The target delay scalar lookup table represents the mapping relationship between different operation features and target delay scalar information. Based on the position information of each operational feature in the first target feature sequence, determine the first target position scalar information corresponding to the first target feature sequence; The first target delay scalar information and the first target position scalar information are concatenated to obtain the fourth target feature sequence; The fourth target feature sequence is input into the encoding layer of the encoder for encoding processing to obtain the third target feature sequence.

2. The method according to claim 1, characterized in that, The decoder includes a first decoding layer and a second decoding layer. The step of inputting the second target feature sequence and the third target feature sequence into the decoder of the operation prediction model for operation decoding processing to obtain the target operation information corresponding to the target object at the time to be predicted includes: Based on the target delay scalar lookup table in the operation prediction model, the delay scalar information corresponding to each operation feature in the second target feature sequence is looked up to obtain the second target delay scalar information corresponding to the second target feature sequence. The target delay scalar lookup table represents the mapping relationship between different operation features and target delay scalar information. Based on the position information of each operational feature in the second target feature sequence, determine the second target position scalar information corresponding to the second target feature sequence; The second target delay scalar information and the second target position scalar information are concatenated to obtain the fifth target feature sequence; The fifth target feature sequence is input into the first decoding layer for decoding processing to obtain the sixth target feature sequence; The sixth target feature sequence and the third target feature sequence are input into the second decoding layer for decoding processing to obtain the target operation information.

3. The method according to claim 1, characterized in that, The method further includes: Obtain the preset prediction model; Before obtaining the first operation feature sequence corresponding to the target object, the method further includes: Obtain the historical operation feature sequence corresponding to the target object. The historical operation feature sequence represents the operation characteristics of the target object within a historical time period, and the historical time period includes the first preset time period. The historical operation feature sequence is segmented to obtain the first operation feature sequence and the second operation feature sequence; Based on the second operation feature sequence, the preset prediction model is trained to obtain the trained operation prediction model.

4. The method according to claim 3, characterized in that, The preset prediction model includes a preset encoder and a preset decoder. The step of training the preset prediction model based on the second operation feature sequence to obtain the trained operation prediction model includes: Based on the second operation feature sequence, at least one set of sample feature sequences and operation information labels corresponding to the at least one set of sample feature sequences are determined, and each set of sample feature sequences includes a first sample feature sequence and a second sample feature sequence. The first sample feature sequence is input into the preset encoder for encoding processing to obtain the third sample feature sequence; The second sample feature sequence and the third sample feature sequence are input into the preset decoder for operation decoding processing to obtain prediction operation information; Based on the predicted operation information and the operation information label, operation loss information is obtained; The preset prediction model is trained based on the operational loss information to obtain the operational prediction model.

5. The method according to claim 4, characterized in that, The step of inputting the first sample feature sequence into the preset encoder for encoding processing to obtain the third sample feature sequence includes: Based on the preset delay scalar lookup table in the preset prediction model, the delay scalar information corresponding to each operation feature in the first sample feature sequence is found to obtain the sample delay scalar information corresponding to the first sample feature sequence. The preset delay scalar lookup table represents the mapping relationship between different operation features and preset delay scalar information. Based on the position information of each operational feature in the first sample feature sequence, determine the sample position scalar information corresponding to the first sample feature sequence; The sample delay scalar information and the sample position scalar information are concatenated to obtain the fourth sample feature sequence; The fourth sample feature sequence is input into the encoding layer of the preset encoder for encoding processing to obtain the third sample feature sequence.

6. The method according to claim 5, characterized in that, The step of training the preset prediction model based on the operational loss information to obtain the operational prediction model includes: Based on the operation loss information, the model parameters in the preset prediction model and the preset delay scalar information in the preset delay scalar lookup table are updated to obtain the updated preset prediction model. Based on the updated preset prediction model, the process of determining at least one set of sample feature sequences and operation information labels corresponding to the at least one set of sample feature sequences based on the second operation feature sequence is repeated. Based on the operation loss information, the model parameters in the preset prediction model and the preset delay scalar information in the preset delay scalar lookup table are updated to obtain the updated preset prediction model. This process continues until a preset convergence condition is reached, and the preset prediction model obtained when the preset convergence condition is reached is taken as the operation prediction model.

7. The method according to claim 4, characterized in that, The method further includes: Obtain the preset time span upper limit and the time span information of the second operation feature sequence; The step of determining at least one set of sample feature sequences and operation information tags corresponding to the at least one set of sample feature sequences based on the second operation feature sequence includes: Based on the time span information of the second operation feature sequence and the preset time span upper limit, the second operation feature sequence is grouped to obtain at least one third operation feature sequence; Based on the target operation feature in each third operation feature sequence, an operation information label is generated corresponding to each third operation feature sequence, wherein the target operation feature is the operation feature with the latest operation time corresponding to each third operation feature sequence; Extract the operation feature sequence other than the target operation feature from each of the third operation feature sequences to obtain the fourth operation feature sequence corresponding to each of the third operation feature sequences; Based on the time span information of each fourth operation feature sequence, the segmentation position information corresponding to each fourth operation feature sequence is determined; Based on the segmentation location information, each fourth operation feature sequence is segmented to obtain the at least one set of sample feature sequences.

8. An operation prediction device, characterized in that, The device includes: The sequence acquisition module is used to acquire a first operation feature sequence corresponding to a target object. The first operation feature sequence represents the operation features of the target object in a first preset time period. The target object is a user account. Different operation types correspond to different operation features. The operation types include keyword search operation, category filtering operation, or consumption operation. The operation feature corresponding to the keyword search operation is the feature corresponding to the searched keyword information. The operation feature corresponding to the category filtering operation is the feature corresponding to the filtered category information. The operation feature corresponding to the consumption operation is the feature corresponding to the consumption information. The preprocessing module is used to segment the first operational feature sequence to obtain a first target feature sequence and a second target feature sequence. The execution module is used to input the first target feature sequence into the encoder in the operation prediction model for encoding processing to obtain the third target feature sequence; An operation prediction module is used to input the second target feature sequence and the third target feature sequence into the decoder in the operation prediction model for operation decoding processing, so as to obtain the target operation information of the target object at the time to be predicted. The execution module is used for: Based on the target delay scalar lookup table in the operation prediction model, the delay scalar information corresponding to each operation feature in the first target feature sequence is looked up to obtain the first target delay scalar information corresponding to the first target feature sequence. The target delay scalar lookup table represents the mapping relationship between different operation features and target delay scalar information. Based on the position information of each operational feature in the first target feature sequence, determine the first target position scalar information corresponding to the first target feature sequence; The first target delay scalar information and the first target position scalar information are concatenated to obtain the fourth target feature sequence; The fourth target feature sequence is input into the encoding layer of the encoder for encoding processing to obtain the third target feature sequence.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to execute the executable instructions to implement the operation prediction method according to any one of claims 1 to 7.

10. A non-volatile computer-readable storage medium storing computer program instructions thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the operation prediction method according to any one of claims 1 to 7.

11. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the operation prediction method according to any one of claims 1 to 7.

Images