Data processing method, data processing model training method and computing device

Abstract

The embodiment of the invention provides a data processing method, a data processing model training method and computing equipment. The data processing method comprises the following steps: receiving user behavior data of a user; the sequence type behavior data and / or the non-sequence type behavior data are / is input into a data processing model, a data processing result corresponding to the user is obtained, the data processing model is obtained through training of sequence type sample features and non-sequence type sample features, and the sequence type sample features and the non-sequence type sample features are obtained through training of the sequence type sample features and the non-sequence type sample features. And the feature enhancement module is obtained by performing feature extraction on a sequence type sample and a non-sequence type sample in the data processing model to obtain an initial sequence type sample feature and an initial non-sequence type sample feature, and performing feature enhancement on the initial sequence type sample feature and the initial non-sequence type sample feature. The problem that the model training efficiency is poor due to sample data is avoided, and the requirements of practical application are met.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of computer technologies, and particularly to a data processing method, a data processing model training method, and a computing device. Background Art

[0002] With the continuous development of computer and artificial intelligence technologies, machine learning has achieved success in multiple fields. Therefore, many neural network models obtained by training based on sample data have emerged.

[0003] In order to improve the performance of neural network models, the prior art often only focuses on the innovation of the neural network model structure, resulting in the model becoming increasingly complex, and thus the model training efficiency is poor and cannot meet the needs of practical applications. Summary of the Invention

[0004] In view of this, the embodiments of this specification provide a data processing method. One or more embodiments of this specification also relate to another data processing method, a data processing model training method, an object recommendation model training method, a data processing device, another data processing device, a data processing model training device, an object recommendation model training device, a computing device, a computer-readable storage medium, and a computer program to solve the technical defects existing in the prior art.

[0005] According to the first aspect of the embodiments of this specification, a data processing method is provided, including:

[0006] Receiving user behavior data, where the user behavior data includes sequential behavior data and / or non-sequential behavior data;

[0007] Inputting the sequential behavior data and / or the non-sequential behavior data into a data processing model to obtain a data processing result corresponding to the user, where the data processing model is obtained by training with sequential sample features and non-sequential sample features, and the sequential sample features and the non-sequential sample features are obtained by performing feature extraction on sequential samples and non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features, and performing feature enhancement on the initial sequential sample features and the initial non-sequential sample features.

[0008] According to the second aspect of the embodiments of this specification, a data processing device is provided, including:

[0009] A data receiving module configured to receive user behavior data, where the user behavior data includes sequential behavior data and / or non-sequential behavior data;

[0010] A data processing module, configured to input the sequential behavior data and / or the non-sequential behavior data into a data processing model to obtain a data processing result corresponding to the user, where the data processing model is obtained by training with sequential sample features and non-sequential sample features, and the sequential sample features and the non-sequential sample features are obtained by extracting features from sequential samples and non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features, and performing feature enhancement on the initial sequential sample features and the initial non-sequential sample features.

[0011] According to a third aspect of the embodiments of the present specification, a method for training a data processing model is provided, including:

[0012] Determine sequential samples and non-sequential samples of a data processing model to be trained, where the sequential samples and the non-sequential samples are user behavior data of sample users;

[0013] Input the sequential samples and the non-sequential samples into the data processing model to be trained, and extract features from the sequential samples and the non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features;

[0014] Perform feature enhancement on the initial sequential sample features to obtain the sequential sample features, and perform feature enhancement on the initial non-sequential sample features to obtain the non-sequential sample features;

[0015] Train the data processing model to be trained based on the sequential sample features and the non-sequential sample features to obtain a data processing model.

[0016] According to a fourth aspect of the embodiments of the present specification, a data processing model training device is provided, including:

[0017] A sample determination module, configured to determine sequential samples and non-sequential samples of a data processing model to be trained, where the sequential samples and the non-sequential samples are user behavior data of sample users;

[0018] A feature extraction module, configured to input the sequential samples and the non-sequential samples into the data processing model to be trained, and extract features from the sequential samples and the non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features;

[0019] A sample enhancement module, configured to enhance the features of the initial sequential sample features to obtain the sequential sample features, and enhance the features of the initial non-sequential sample features to obtain the non-sequential sample features;

[0020] A model training module, configured to train the data processing model to be trained based on the sequential sample features and the non-sequential sample features to obtain a data processing model.

[0021] According to the fifth aspect of the embodiments of the present specification, another data processing method is provided, which is applied to the cloud and includes:

[0022] Receiving user behavior data sent by a terminal, where the user behavior data includes sequential behavior data and / or non-sequential behavior data;

[0023] Inputting the sequential behavior data and / or the non-sequential behavior data into a data processing model to obtain a data processing result corresponding to the user, where the data processing model is obtained by training with sequential sample features and non-sequential sample features, and the sequential sample features and the non-sequential sample features are obtained by extracting features from sequential samples and non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features, and enhancing the initial sequential sample features and the initial non-sequential sample features;

[0024] Sending the data processing result to the terminal.

[0025] According to the sixth aspect of the embodiments of the present specification, another data processing device is provided, which is applied to the cloud and includes:

[0026] A data receiving module, configured to receive user behavior data sent by a terminal, where the user behavior data includes sequential behavior data and / or non-sequential behavior data;

[0027] A data processing module, configured to input the sequential behavior data and / or the non-sequential behavior data into a data processing model to obtain a data processing result corresponding to the user, where the data processing model is obtained by training with sequential sample features and non-sequential sample features, and the sequential sample features and the non-sequential sample features are obtained by extracting features from sequential samples and non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features, and enhancing the initial sequential sample features and the initial non-sequential sample features;

[0028] A result sending module, configured to send the data processing result to the terminal.

[0029] According to a seventh aspect of the embodiments of the present specification, there is provided a method for training an object recommendation model, including:

[0030] Determine a sequential sample and a non-sequential sample of the object recommendation model to be trained, where the sequential sample and the non-sequential sample are user behavior data of a sample user for a sample object;

[0031] Input the sequential sample and the non-sequential sample into the object recommendation model to be trained, and perform feature extraction on the sequential sample and the non-sequential sample in the object recommendation model to obtain an initial sequential sample feature and an initial non-sequential sample feature;

[0032] Perform feature enhancement on the initial sequential sample feature to obtain the sequential sample feature, and perform feature enhancement on the initial non-sequential sample feature to obtain the non-sequential sample feature;

[0033] Based on the sequential sample feature and the non-sequential sample feature, train the object recommendation model to be trained to obtain an object recommendation model.

[0034] According to an eighth aspect of the embodiments of the present specification, there is provided an object recommendation model training device, including:

[0035] A sample determination module configured to determine a sequential sample and a non-sequential sample of the object recommendation model to be trained, where the sequential sample and the non-sequential sample are user behavior data of a sample user for a sample object;

[0036] A feature extraction module configured to input the sequential sample and the non-sequential sample into the object recommendation model to be trained, and perform feature extraction on the sequential sample and the non-sequential sample in the object recommendation model to obtain an initial sequential sample feature and an initial non-sequential sample feature;

[0037] A sample enhancement module configured to perform feature enhancement on the initial sequential sample feature to obtain the sequential sample feature, and perform feature enhancement on the initial non-sequential sample feature to obtain the non-sequential sample feature;

[0038] A model training module configured to train the object recommendation model to be trained based on the sequential sample feature and the non-sequential sample feature to obtain an object recommendation model.

[0039] According to a ninth aspect of the embodiments of the present specification, there is provided a computing device, including:

[0040] A memory and a processor;

[0041] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the above two data processing methods, data processing model training methods, or object recommendation model training methods are implemented.

[0042] According to the tenth aspect of the embodiments of the present specification, a computer-readable storage medium is provided, which stores computer-executable instructions. When the instructions are executed by a processor, the steps of the above two data processing methods, data processing model training methods, or object recommendation model training methods are implemented.

[0043] According to the eleventh aspect of the embodiments of the present specification, a computer program is provided. When the computer program is executed on a computer, the computer is made to execute the steps of the above two data processing methods, data processing model training methods, or object recommendation model training methods.

[0044] One or more embodiments of the present specification provide a data processing method, including: receiving user behavior data of a user, where the user behavior data includes sequential behavior data and / or non-sequential behavior data; inputting the sequential behavior data and / or the non-sequential behavior data into a data processing model to obtain a data processing result corresponding to the user, where the data processing model is obtained by training with sequential sample features and non-sequential sample features, and the sequential sample features and the non-sequential sample features are obtained by extracting features from sequential samples and non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features, and performing feature enhancement on the initial sequential sample features and the initial non-sequential sample features.

[0045] Specifically, the data processing model in the data processing method is obtained by training with sequential sample features and non-sequential sample features. The sequential sample features and non-sequential sample features are obtained by performing feature extraction on sequential samples and non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features, and then performing feature enhancement; thereby fully considering the influence of sample data on model training during the model training process, avoiding the problem of poor model training efficiency caused by sample data; and, processing the sequential behavior data and / or non-sequential behavior data included in the user behavior data through the trained data processing model, and obtaining a data processing result, thereby meeting the needs of practical applications through the data processing model. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1It is a schematic diagram of the application of a data processing method provided by an embodiment of this specification;

[0047] Figure 2 It is a flowchart of a data processing method provided by an embodiment of this specification;

[0048] Figure 3 It is a model schematic diagram of a data processing model training method provided by an embodiment of this specification;

[0049] Figure 4 It is a flowchart of a data processing model training method provided by an embodiment of this specification;

[0050] Figure 5 It is a flowchart of constructing a co-occurrence matrix in a data processing model training method provided by an embodiment of this specification;

[0051] Figure 6 It is a schematic diagram of data augmentation in a data processing model training method provided by an embodiment of this specification;

[0052] Figure 7 It is a schematic diagram of training a sequence-based contrast learning model in a data processing model training method provided by an embodiment of this specification;

[0053] Figure 8 It is a schematic diagram of sequence feature data enhancement in a data processing model training method provided by an embodiment of this specification;

[0054] Figure 9 It is a schematic diagram of two feature masking methods in a data processing model training method provided by an embodiment of this specification;

[0055] Figure 10 It is a schematic diagram of training a non-sequence-based contrast learning model in a data processing model training method provided by an embodiment of this specification;

[0056] Figure 11 It is a flowchart of the processing process of a data processing model training method provided by an embodiment of this specification;

[0057] Figure 12 It is a flowchart of another data processing method provided by an embodiment of this specification;

[0058] Figure 13 It is a flowchart of an object recommendation model training method provided by an embodiment of this specification;

[0059] Figure 14 It is a structural block diagram of a computing device provided by an embodiment of this specification. Detailed implementation manners

[0060] Numerous specific details are set forth in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the connotation of this specification. Therefore, this specification is not limited by the specific implementations disclosed below.

[0061] The terms used in one or more embodiments of this specification are for the purpose of describing specific embodiments only and are not intended to limit one or more embodiments of this specification. The singular forms "a", "the", and "said" used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0062] It should be understood that although the terms first, second, etc. may be used in one or more embodiments of this specification to describe various information, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "when" or "while" or "in response to determining".

[0063] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data that have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of the relevant data need to comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding operation entrances are provided for the user to choose to authorize or reject.

[0064] In one or more embodiments of this specification, a large model refers to a deep learning model with a large number of model parameters, usually including hundreds of millions, tens of billions, hundreds of billions, trillions or even more than one quadrillion model parameters. A large model can also be referred to as a Foundation Model. Through pre-training of a large model with a large amount of unlabeled corpus, a pre-trained model with more than one hundred million parameters is produced. Such a model can adapt to a wide range of downstream tasks and has good generalization ability. For example, large language models (LLMs), multi-modal pre-training models, etc.

[0065] When a large model is actually applied, only a small number of samples are needed to fine-tune the pre-trained model for application to different tasks. Large models can be widely applied in fields such as natural language processing (NLP), computer vision, etc. Specifically, they can be applied to tasks in the field of computer vision such as visual question answering (VQA), image captioning (IC), image generation, etc., and tasks in the field of natural language processing such as text-based sentiment classification, text summary generation, machine translation, etc. The main application scenarios of large models include digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, etc.

[0066] First, the noun terms involved in one or more embodiments of this specification are explained.

[0067] CL4SRec: Short for Contrastive Learning for Sequential Recommendation, it is a contrastive learning model that only considers sequential features in the recommendation scenario.

[0068] CL4CTR: Short for A Contrastive Learning Framework for CTR Prediction, it is a contrastive learning model that only considers non-sequential features in the recommendation scenario.

[0069] Transformer: A deep learning model used for natural language processing (NLP) tasks, and is now mostly used in the encoder of the model.

[0070] MaskNet: Short for Introducing Feature-Wise Multiplication to CTR Ranking Models by Instance-Guided Mask, it is a CTR model that can better mine cross features.

[0071] CTR: The CTR (Click-Through Rate) model is a machine learning model used to predict whether a user will click on a certain page or link.

[0072] EasyRec: A machine learning framework designed for the recommendation scenario, which has implemented various machine learning models for common recommendation tasks.

[0073] NCE Loss: Short for Noise-Contrastive Estimation Loss, it is a loss function used for training classification models, etc.

[0074] BERT model: A language representation model. BERT stands for Bidirectional Encoder Representations from Transformers. BERT aims to pre-train deep bidirectional representations by jointly conditioning the left and right contexts in all layers.

[0075] Duorec model: Short for Contrastive Learning for Representation Degeneration Problem in Sequential Recommendation. It is a contrastive learning recommendation model that improves CL4SRec and only considers sequential data.

[0076] DNN: Refers to the deep neural network algorithm.

[0077] With the continuous development of computer and artificial intelligence technology, machine learning has achieved success in many fields, so many neural network models based on sample data training have emerged. In order to improve the performance of neural network models, the model training process often only focuses on the innovation of neural network model structure, which makes the model gradually complicated, but does not take into account the problems of sample data itself, resulting in poor model training efficiency and failure to meet the needs of practical applications. For example, in one or more embodiments provided in this specification, with the success of machine learning in many fields, researchers have built many modern recommendation systems based on highly expressive deep neural architectures. A recommendation system is an intelligent system that can recommend personalized content to users, mainly by analyzing the user's historical behavior and interests to predict their future needs, simplifying the decision-making process for users and providing better service experience. However, these recommendation models built based on deep networks often have too high data requirements and require a large amount of data to train the model. In the recommendation system, personalized recommendations rely on the data generated by the user himself, which leads to a high cost of data acquisition; and most users interact (such as consume or click) with only a small part of the many item sets. Therefore, the sparsity of data has become one of the main reasons that affect the full potential of recommendation models. Although the diversity and sparsity of data in recommendation scenarios often limit the performance of recommendation models, and there may be missing or too few samples in actual application scenarios, resulting in poor model training efficiency, researchers often only focus on the innovation of model structure, which leads to the gradual complexity of the model, while ignoring the problems with the data itself.

[0078] In view of the above problems, although contrastive learning algorithms have been proven to effectively solve the problem of data sparsity in many fields, most of the research on contrastive learning algorithms in the recommendation field is carried out on time series features, lacking consideration of non-sequential features. At the same time, researchers pay too much attention to innovations in model structure, resulting in increasingly complex models and high training costs, which cannot be implemented in actual application scenarios. For example, this manual provides two models, the first model is the CL4SRec model; however, since models such as CL4SRec are contrastive learning models that only consider sequence features in recommendation scenarios, they only consider sequence data, and the features are single, which makes it difficult to apply in actual scenarios. The second model is the CL4CTR model; however, models such as CL4CTR are contrastive learning models that only consider non-sequential features in recommendation scenarios. It only considers non-sequential data, and the training speed cost is high, and it is also difficult to migrate to actual scenarios. Based on this, the data in the recommendation scenario has sparsity and long tail problems, as well as how to efficiently apply self-supervision and contrastive learning algorithms in recommendation scenarios, which have become problems that need to be solved urgently.

[0079] In this specification, a data processing method is provided. This specification also relates to another data processing method, a data processing model training method, an object recommendation model training method, a data processing device, another data processing device, a data processing model training device, an object recommendation model training device, a computing device, and a computer-readable storage medium, which will be described in detail one by one in the following embodiments.

[0080] Refer to Figure 1 , Figure 1 FIG. shows an application schematic diagram of a data processing method provided according to an embodiment of this specification, and is described in the case where the data processing method provided in this specification is applied to a product recommendation scenario. Among them, user behavior data can be understood as data generated during the process of a user's operation behaviors such as browsing a product page, purchasing a product, and favoriting a product for a product. The user behavior data includes sequential behavior data and / or non-sequential behavior data; the sequential behavior data can be an interaction record between the user and the product. For example, the interaction record can consist of three time series features of product ID, product category ID, and user behavior sequence. The non-sequential behavior data can be user attribute data such as user age and user gender, and product attribute data such as product price and product capacity. Based on this, refer to Figure 1 , the user provides their user behavior data to the server 104 through the terminal 102, and the user behavior data includes sequential behavior data and / or non-sequential behavior data. After receiving the sequential behavior data and / or non-sequential behavior data, the server 104 inputs it into the product recommendation model for processing, so as to obtain a product recommendation result predicted for the user. Subsequently, product recommendations can be made for the user according to the product recommendation result, thus meeting the needs of actual applications and providing a good shopping experience for the user.

[0081] Refer to Figure 2 , Figure 2 FIG. shows a flowchart of a data processing method provided according to an embodiment of this specification, which specifically includes the following steps.

[0082] Step 202: Receive the user's user behavior data, where the user behavior data includes sequential behavior data and / or non-sequential behavior data.

[0083] Among them, the user behavior data can be understood as the data generated during the user's operation behavior; when the data processing method provided in the embodiments of this specification is applied to different scenarios, the operation behavior and the user behavior data are also different. For example, when the data processing method is applied to the product recommendation scenario, the operation behavior can be understood as the operation behaviors of the user browsing the product page, purchasing the product, collecting the product, etc. for the product. The user behavior data can be understood as the data generated during the operation behaviors of the user browsing the product page, purchasing the product, collecting the product, etc. for the product. It should be noted that the user can perform the above various operation behaviors on the product through the Internet shopping platform; the user behavior data can be the user behavior data of the user stored in the Internet shopping platform. For another example, when the data processing method is applied to the scenic spot recommendation scenario, the operation behavior can be understood as the operation behaviors of the user browsing the scenic spot page, purchasing the scenic spot ticket, collecting the scenic spot, etc. for the scenic spot. The user behavior data can be understood as the data generated during the operation behaviors of the user browsing the scenic spot page, purchasing the scenic spot ticket, collecting the scenic spot, etc. for the scenic spot. It should be noted that the user can perform the above various operation behaviors on the scenic spot through the Internet scenic spot ticket purchasing platform; the user behavior data can be the user behavior data of the user stored in the Internet scenic spot ticket purchasing platform.

[0084] The sequential behavior data can be understood as the behavior data sorted into a sequence according to specific parameters. For example, the sequential behavior data can be time-series behavior data, and the time-series behavior data can be the behavior data sorted according to time. In one or more embodiments provided in this specification, the sequential behavior data can be sequential behavior features, and the sequential behavior features can be the interaction records between the user and the product. For example, the interaction record can be composed of three time-series features of product ID, product category ID, and user behavior sequence, where different time-series features are in one-to-one correspondence within the same time period.

[0085] The non-sequential behavior data can be understood as the behavior data that is not sorted into a sequence according to specific parameters. For example, the non-sequential behavior data can be user attribute data such as user age and user gender, and product attribute data such as product price and product capacity. In one or more embodiments provided in this specification, the non-sequential behavior data can be non-sequential behavior features; for example, the non-sequential behavior features can be user attribute features such as user age and user gender, and product attribute features such as product price and product capacity.

[0086] In one or more embodiments provided in this specification, the data processing method further includes receiving user behavior data of a user and determining that the user behavior data includes sequential behavior data and / or non-sequential behavior data; extracting features from the sequential behavior data and / or the non-sequential behavior data included in the user behavior data to obtain sequential behavior features and / or non-sequential behavior features; thereby avoiding the problem of a large model caused by configuring a network layer for feature extraction in the data processing model, and enabling the data processing model to be compatible with more hardware devices. Subsequently, the sequential behavior features and / or the non-sequential behavior features are input into the data processing model to obtain the data processing result corresponding to the user. For example, in the data processing method provided in this specification, the user behavior data may be a data set; after receiving the data set provided by the user, sequential behavior data and / or non-sequential behavior data are determined from the data set; then, features are extracted from the sequential behavior data and / or the non-sequential behavior data to obtain data features; and then, according to the feature types of the data features, the data features are divided into sequential behavior features and non-sequential behavior features.

[0087] Step 204: Input the sequential behavior data and / or the non-sequential behavior data into a data processing model to obtain the data processing result corresponding to the user.

[0088] Among them, the data processing model is obtained by training with sequential sample features and non-sequential sample features. The sequential sample features and the non-sequential sample features are obtained by extracting features from sequential samples and non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features, and performing feature enhancement on the initial sequential sample features and the initial non-sequential sample features.

[0089] The data processing model can be understood as a neural network model capable of processing the sequential behavior data and / or non-sequential behavior data; when the data processing method provided in the embodiments of this specification is applied to different scenarios, the data processing model is also different. For example, when the data processing method is applied to the product recommendation scenario, the data processing model can be a product recommendation model; the product recommendation model can process the user behavior data generated during the user's various operations on products and output corresponding product recommendation results for the user's corresponding terminal. Subsequently, product recommendations can be made for the user through the product recommendation results. Another example is that when the data processing method is applied to the scenic spot recommendation scenario, the data processing model can be a scenic spot recommendation model; the scenic spot recommendation model can process the user behavior data generated during the user's various operations on scenic spots and output corresponding scenic spot recommendation results for the user. Subsequently, scenic spot recommendations can be made for the user through the scenic spot recommendation results. Based on this, the data processing model can be understood as an object recommendation model; the object recommendation model can process the sequential behavior data and / or non-sequential behavior data to obtain the user's object recommendation results.

[0090] In one or more embodiments provided in this specification, the data processing model includes a first data processing module and a second data processing module; wherein, the first data processing module can extract features from the sequential behavior data to obtain the sequential behavior features of the sequential behavior data, and use an encoder to perform encoding processing on the sequential behavior features to obtain the object recommendation results determined by the sequential behavior data. Among them, the second data processing module can extract features from the non-sequential behavior data to obtain the non-sequential behavior features of the non-sequential behavior data, and use an encoder to perform encoding processing on the non-sequential behavior features to obtain the object recommendation results determined by the non-sequential behavior data. It should be noted that when the sequential behavior data or non-sequential behavior data is input into the data processing model, the data processing model can use the first data processing module or the second data processing module for processing to obtain the user's object recommendation results. When the sequential behavior data and non-sequential behavior data are input into the data processing model, the data processing model can use the first data processing module and the second data processing module for processing, and regard the object recommendation results output by the first data processing module and the second data processing module as the user's object recommendation results.

[0091] The data processing result can be understood as the result of the data processing model processing the sequential behavior data and / or the non-sequential behavior data. When the data processing method provided in the embodiment of this specification is applied to different scenarios, the data processing result is also different. For example, when the data processing method is applied to the commodity recommendation scenario, the data processing result is the commodity recommendation result, and the commodity recommendation result can be the commodity information of the commodity to be recommended to the user, for example, the commodity information can include: commodity type: skin care products, commodity price: 100 to 150 yuan, commodity brand, commodity capacity and other information. Subsequently, commodity recommendations can be made to users through commodity information. For another example, when the data processing method is applied to the scenic spot recommendation scenario, the data processing result is the scenic spot recommendation result, and the scenic spot recommendation result can be the scenic spot information of the scenic spot to be recommended to the user, for example, the scenic spot information can include: scenic spot type: farmhouse, scenic spot price: 100 to 150 yuan, scenic spot region, scenic spot name and other information. Subsequently, scenic spot recommendations can be made to users through scenic spot information.

[0092] The sequence sample can be understood as the sequence behavior data in the user behavior data of the sample user, that is, the sequence sample can be understood as the sequence behavior data as a training sample. For example, the sequence sample can be an interaction record between a user and a product as a training sample, and the interaction record can be composed of three time series features consisting of a product ID, a product category ID, and a user behavior sequence, wherein different time series features are one-to-one corresponding in the same time period. The non-sequential sample can be understood as the sequence behavior data in the user behavior data of the sample user, that is, the non-sequential sample can be understood as the non-sequential behavior data as a training sample. For example, user attribute data such as user age and user gender as training samples, and product attribute data such as product price and product capacity. In one or more embodiments provided in this specification, the non-sequential behavior data can be a non-sequential behavior feature; for example, the non-sequential behavior feature can be a user attribute feature such as user age and user gender, and a product attribute feature such as product price and product capacity. The non-sequential sample features can be understood as sample features extracted from non-sequential samples and feature-enhanced; the non-sequential sample features are used to train the data processing model. The sequence sample features can be understood as sample features extracted from sequence samples and feature-enhanced; the sequence sample features are used to train the data processing model.

[0093] Specifically, in the data processing method according to the embodiments of this specification, after receiving the user behavior data of a user, the sequential behavior data and / or non-sequential behavior data will be determined from the user behavior data; then the sequential behavior data and / or non-sequential behavior data will be input into a data processing model, so as to obtain the data processing result corresponding to the user. In one or more embodiments provided in this specification, the step of inputting the sequential behavior data and / or the non-sequential behavior data into the data processing model to obtain the data processing result corresponding to the user includes: inputting the sequential behavior data and / or the non-sequential behavior data into the data processing model; using a first data processing module in the data processing model to extract features from the sequential behavior data to obtain sequential behavior features of the sequential behavior data, and using an encoder in the first data processing module to perform encoding processing on the sequential behavior features to obtain a first data processing result, and / or using a second data processing module in the data processing model to extract features from the non-sequential behavior data to obtain non-sequential behavior features of the non-sequential behavior data, and using an encoder in the second data processing module to perform encoding processing on the non-sequential behavior features to obtain a second data processing result; determining the data processing result corresponding to the user based on the first data processing result and / or the second data processing result.

[0094] The data processing model in the data processing method provided in this specification can process sequential behavior data through a first data processing module and / or process non-sequential behavior data through a second data processing module, thereby fully considering the influence of sequential behavior data and non-sequential behavior data on the data processing model to determine the data processing result, so as to obtain a more accurate data processing result and meet the requirements of the actual application scenario.

[0095] In one or more embodiments provided in this specification, after determining the data processing result corresponding to the user, the data processing result can be sent to the user.

[0096] In one or more embodiments provided in this specification, in response to problems in the model training process, the data processing method provided in this specification proposes a data processing model for full-feature contrast learning, which can consider both sequential data features and non-sequential data features. In addition, the prediction of the model training speed and cost is also improved, so as to optimize the model from the perspectives of training time and training cost. While improving the model prediction effect, it provides help for the prediction of actual recommendation scenarios and meets the requirements of actual application scenarios. Specifically, in the training process of this data processing model, the data processing method provided in this specification takes into account that existing related work often focuses on the innovation of the model structure and ignores optimizing the model prediction results from the data perspective. For example, contrast learning models in recommendation scenarios often study sequential features and non-sequential features separately, which does not conform to the actual application scenario. Based on this, the data processing method provided in this specification provides a self-supervised dataset augmentation method to optimize the model prediction effect from the sample data perspective. In addition, in terms of the model structure, the data processing method provided in the embodiments of this specification can simultaneously provide a contrast learning model for sequential features and non-sequential features, thus meeting the requirements of actual application scenarios. The specific training process for this data processing model is as follows: Before inputting the sequential behavior data and / or the non-sequential behavior data into the data processing model to obtain the data processing result corresponding to the user, it further includes:

[0097] Determine the sequential samples and the non-sequential samples of the data processing model to be trained, where the sequential samples and the non-sequential samples are the user behavior data of sample users;

[0098] Input the sequential samples and the non-sequential samples into the data processing model to be trained, and perform feature extraction on the sequential samples and the non-sequential samples in the data processing model to be trained to obtain initial sequential sample features and initial non-sequential sample features;

[0099] Perform feature enhancement on the initial sequential sample features to obtain the sequential sample features, and perform feature enhancement on the initial non-sequential sample features to obtain the non-sequential sample features;

[0100] Train the data processing model to be trained based on the sequential sample features and the non-sequential sample features to obtain the data processing model.

[0101] Among them, the data processing model to be trained can be understood as the data processing model that needs to be trained;

[0102] Specifically, before the data processing method provided in this specification inputs the sequential behavior data and / or the non-sequential behavior data into the data processing model to obtain the data processing result corresponding to the user, the data processing model needs to be trained first. The training steps can be as follows: Determine the sequential samples and non-sequential samples required for model training for the data processing model to be trained; among them, the sequential samples and non-sequential samples are the user behavior data of the sample users; that is to say, the sequential behavior data included in the user behavior data of the sample users is used as the sequential samples, and the non-sequential behavior data is used as the non-sequential samples.

[0103] After obtaining the sequential samples and non-sequential samples, the sequential samples and non-sequential samples are input into the data processing model to be trained. In this data processing model to be trained, feature extraction is performed on the sequential samples to obtain initial sequential sample features, and feature extraction is performed on the non-sequential samples to obtain initial non-sequential sample features.

[0104] After obtaining the initial sequential sample features and initial non-sequential sample features, feature enhancement needs to be performed on the initial sequential sample features and initial non-sequential sample features. For example, by using methods such as vector masking operation, vector clipping operation, or vector reordering operation, feature enhancement is performed on the initial sequential sample features to obtain sequential sample features. Feature enhancement is performed on the initial non-sequential sample features through feature masking to obtain non-sequential sample features.

[0105] After obtaining the sequential sample features and non-sequential sample features, the data processing model to be trained is trained using the sequential sample features and non-sequential sample features until the model training stop condition is reached, and a trained data processing model is obtained. Subsequently, the sequential behavior data and / or non-sequential behavior data are processed based on this data processing model, thus meeting the needs of practical applications.

[0106] It should be noted that the above are the training steps for the data processing model to be trained in this embodiment. These training steps are within the same concept as the steps in the data processing model training method in one or more embodiments of this specification. For the detailed content not described in detail in these training steps, reference can be made to the corresponding descriptions in one or more embodiments of this specification, and no further elaboration will be provided here.

[0107] In one or more embodiments provided in this specification, during the process of model training, the data processing method provided in this specification takes into account that the number of samples has a great impact on the model training efficiency. To improve the performance of the data processing model after training, the data processing method provided in this specification will augment the sample data, thereby increasing the number of samples and further ensuring the performance of the data processing model after training. The specific method is as follows.

[0108] The feature enhancement of the initial sequential sample feature to obtain the sequential sample feature includes:

[0109] Determine a similar behavior object corresponding to the user behavior object in the initial sequential sample feature, and perform sample augmentation based on the user behavior object and the similar behavior object to obtain an augmented sequential sample feature;

[0110] Perform feature enhancement on the initial sequential sample feature and the augmented sequential sample feature to obtain the sequential sample feature of the data processing model to be trained.

[0111] Among them, the user behavior object can be understood as the behavior object targeted by the user behavior. For example, when the data processing method is applied to the commodity recommendation scenario, the user behavior object can be understood as the target commodity corresponding to the user behavior, that is, the commodity corresponding to the user's operations such as browsing commodities, purchasing commodities, and collecting commodities. For another example, when the data processing method is applied to the scenic spot recommendation scenario, the user behavior object can be understood as the target scenic spot corresponding to the user behavior, that is, the scenic spot corresponding to the user's operations such as browsing scenic spots, purchasing scenic spot tickets, and collecting scenic spots. For another example, when the data processing method is applied to the food recommendation scenario, the user behavior object can be understood as the target food corresponding to the user behavior, that is, the food corresponding to the user's operations such as food browsing, food purchasing, and food collection. The similar behavior object can be understood as a behavior object that is relatively similar to the user behavior object; for example, when the user behavior object is a facial cleanser of brand A, the similar behavior object can be a facial cleanser of brand B.

[0112] Specifically, after extracting features from sequential samples in the data processing model to be trained and obtaining the initial sequential sample features, the initial sequential sample features can be augmented. First, it is necessary to determine the user behavior objects in the initial sequential sample features. There can be at least two user behavior objects, and corresponding similar behavior objects are determined for each user behavior object based on the similarity results between the user behavior objects. Then, sample augmentation operations are performed based on the user behavior objects and the similar behavior objects to obtain augmented sequential sample features. Based on this, by determining the initial sequential sample features and the augmented sequential sample features as the sample features of the data processing model to be trained, the samples of the data processing model to be trained are augmented, and the number of training samples is increased.

[0113] After obtaining the initial sequential sample features and the augmented sequential sample features, feature enhancement is performed on the initial sequential sample features and the augmented sequential sample features, and the enhanced initial sequential sample features and the enhanced augmented sequential sample features are used as the sequential sample features.

[0114] The data processing model in the data processing method provided by one or more embodiments of this specification is trained using sequential sample features and non-sequential sample features. The sequential sample features and non-sequential sample features are obtained by performing feature enhancement on the initial sequential sample features and the initial non-sequential sample features obtained by extracting features from sequential samples and non-sequential samples in the data processing model. Thus, during the model training process, the influence of sample data on model training is fully considered, and the problem of poor model training efficiency caused by sample data is avoided. Moreover, the trained data processing model processes the sequential behavior data and / or non-sequential behavior data included in the user behavior data and obtains a data processing result, thereby meeting the actual application needs through the data processing model.

[0115] See Figure 3 , Figure 3The figure shows a schematic diagram of the model structure in a data processing model training method provided according to an embodiment of this specification. Among them, the data processing model provided in this specification is a full-feature contrast learning model, and this full-feature contrast learning model is a multi-task learning model composed of a sequential feature contrast learning model, a non-sequential feature contrast learning model, and a multi-layer perceptron model. The specific training process includes extracting features from the original data (such as sequential samples and non-sequential samples) provided by the user to obtain sequential features and non-sequential features. Input the non-sequential features into the non-sequential feature contrast learning model to obtain the model processing result, and calculate the non-sequential feature contrast learning model loss value based on this model processing result. Input the non-sequential features into the embedding layer in the multi-layer perceptron model, obtain the non-sequential feature vector of the non-sequential features, input the non-sequential feature vector into the multi-layer perceptron in the multi-layer perceptron model to obtain the processing result, and calculate the multi-layer perceptron model loss value based on this processing result. Perform data augmentation on the sequential features, and input the augmented sequential features into the sequential feature contrast learning model for processing to obtain the model processing result, and calculate the sequential feature contrast learning model loss value based on this model processing result. Train the model through the above three loss values until the model training stop condition is reached, so as to fully consider the impact of sample data on model training during the model training process, and avoid the problem of poor model training efficiency caused by sample data.

[0116] See Figure 4 , Figure 4 The figure shows a flowchart of a data processing model training method provided according to an embodiment of this specification, which specifically includes the following steps.

[0117] Step 402: Determine the sequential samples and non-sequential samples of the data processing model to be trained, where the sequential samples and the non-sequential samples are the user behavior data of the sample user.

[0118] In the data processing model training method provided in this specification, the user behavior data of the sample user can be a data set, which contains sequential behavior data and non-sequential behavior data; after obtaining the data set provided by the sample user, determine the sequential behavior data and non-sequential behavior data from the data set; then perform feature extraction on the sequential behavior data and non-sequential behavior data as training samples to obtain sequential sample features and non-sequential sample features.

[0119] Step 404: Input the sequential samples and the non-sequential samples into the data processing model to be trained, and perform feature extraction on the sequential samples and the non-sequential samples in the data processing model to be trained to obtain initial sequential sample features and initial non-sequential sample features.

[0120] In one or more embodiments provided in this specification, the sequential sample may be sequential behavior data as a training sample, and the non-sequential sample may be non-sequential behavior data as a training sample. After determining the sequential sample and the non-sequential sample, the sequential sample and the non-sequential sample are input into the data processing model to be trained. Using the feature extraction module in the data processing model to be trained, feature extraction is performed on the sequential sample to obtain initial sequential sample features, and feature extraction is performed on the non-sequential sample to obtain initial non-sequential sample features. Among them, the feature extraction module can be understood as the module in the data processing model to be trained for feature extraction. The feature extraction module can be one or more network layers, or the feature extraction module can be a neural network model in the data processing model to be trained for feature extraction. Subsequently, the initial sequential behavior sample features and the initial non-sequential behavior sample features are used to train the data processing model to be trained.

[0121] Step 406: Perform feature enhancement on the initial sequential sample features to obtain the sequential sample features, and perform feature enhancement on the initial non-sequential sample features to obtain the non-sequential sample features.

[0122] Performing feature enhancement on the initial sequential sample features to obtain the sequential sample features includes

[0123] Determine similar behavior objects corresponding to the user behavior objects in the initial sequential sample features, and perform sample augmentation based on the user behavior objects and the similar behavior objects to obtain augmented sequential sample features;

[0124] Perform feature enhancement on the initial sequential sample features and the augmented sequential sample features to obtain the sequential sample features of the data processing model to be trained.

[0125] Among them, the augmented sequential sample features can be understood as sample data obtained by performing sample augmentation through the user behavior elements in the initial sequential sample features and the similar behavior elements corresponding to the user behavior elements.

[0126] Specifically, after extracting features from sequential samples in the data processing model to be trained and obtaining the initial sequential sample features, the initial sequential sample features can be augmented. First, at least two user behavior objects in the initial sequential sample features need to be determined, and corresponding similar behavior objects are determined for each user behavior object based on the similarity results between the at least two user behavior objects. Then, a sample augmentation operation is performed based on the user behavior objects and the similar behavior objects to obtain augmented sequential sample features. After obtaining the initial sequential sample features and the augmented sequential sample features, feature enhancement is performed on the initial sequential sample features and the augmented sequential sample features, and the enhanced initial sequential sample features and the enhanced augmented sequential sample features are used as sequential sample features. Thereby, the samples of the data processing model to be trained are augmented, and the number of training samples is increased.

[0127] It should be noted that in one or more embodiments provided in this specification, feature enhancement can be performed on the initial sequential sample features and the augmented sequential sample features. Among them, the steps of performing feature enhancement on the initial sequential sample features and the augmented sequential sample features are the same. For the steps of performing feature enhancement on the augmented sequential sample features, reference can be made to the content of performing feature enhancement on the initial sequential sample features to avoid excessive repetition.

[0128] In one or more embodiments provided in this specification, due to problems such as data diversity and sparsity in some application scenarios, there will be a situation where sample data is missing or too few in the actual application scenario, which often limits the performance of the data processing model. Therefore, for the situation of too few data samples, the data processing model training method in one or more embodiments of this specification proposes a self-supervised data augmentation scheme. This scheme is a scheme for augmenting the original data set based on a self-supervised augmentation method, which can augment the original data set from the perspective of data to optimize the prediction effect of the model. Specifically, in the process of augmenting sample data, it is necessary to determine the similar behavior objects corresponding to the user behavior objects in the initial sequential sample features, and use the user behavior objects and the similar behavior objects to perform sample augmentation to obtain augmented sequential sample features. The specific method is as follows. The determination of the similar behavior corresponding to the user behavior in the initial sequence includes steps one to three:

[0129] Step one: Determine at least two user behavior objects in the initial sequential sample features.

[0130] It should be noted that each initial sequential sample feature contains at least two user behavior objects.

[0131] Step 2: Determine the similarity results between the target behavior object and other behavior objects among the at least two user behavior objects, where the target behavior object is any one of the at least two user behavior objects, and the other behavior objects are the other user behavior objects among the at least two user behavior objects except the target behavior object.

[0132] Specifically, the determining the similarity results between the target behavior object and other behavior objects among the at least two user behavior objects includes:

[0133] Construct a user behavior object matrix based on the at least two user behavior objects;

[0134] Obtain a similarity result matrix corresponding to the at least two user behavior objects by performing similarity calculation on the user behavior object matrix;

[0135] Determine the similarity results between the target behavior object and other behavior objects among the at least two user behavior objects through the similarity result matrix.

[0136] Among them, the user behavior object matrix can be understood as a matrix composed of user behavior object matrices; taking the user behavior object as the commodity corresponding to the user behavior as an example, in the process of calculating the similarity between commodities, first, a co-occurrence matrix of commodities needs to be created. It should be noted that there are at least two sample users in this specification, and each sample user has corresponding initial sequence-type sample features and non-sequence-type sample features. The co-occurrence matrix is composed of the commodities in the initial sequence-type sample features of multiple sample users. Then, for each sample user, the commodities in the commodity list interacted by each sample user are added by 1 pairwise in the co-occurrence matrix, so as to adjust the co-occurrence matrix and obtain the adjusted co-occurrence matrix, which can be understood as the user behavior object matrix. For example, refer to Figure 5 , Figure 5 shows a flowchart of constructing a co-occurrence matrix in a data processing model training method provided according to an embodiment of this specification. Among them Figure 5 the A, B, and D included in the commodity list can be understood as different commodities, that is, user behavior objects; Figure 5A, B, C, D, and E contained in the co-occurrence matrix can be understood as different commodities. Based on this, in the process of calculating the similarity of commodities, it is first necessary to determine the commodities contained in each initial sequence-type sample feature, and establish a commodity list based on this commodity (that is, based on the initial sequence-type sample features of each user, establish a list for each user that contains the commodities he likes); secondly, for each user, add 1 to the co-occurrence matrix for each pair of commodities in his commodity list. Specifically, taking commodity category 1 as an example, this commodity category 1 contains three commodities A, B, and D. Match the three commodities in pairs, add the value 1 to the intersection position of commodities A and B in co-occurrence matrix 1; add the value 1 to the intersection position of commodities A and D in co-occurrence matrix 1; add the value 1 to the intersection position of commodities B and D in co-occurrence matrix 1, thus completing the operation of adding 1 to each pair of commodities in commodity list 1 in the co-occurrence matrix, and determining the corresponding co-occurrence matrix 1 for this commodity list 1. In the subsequent similarity calculation process, these co-occurrence matrices can be added together to obtain a target matrix that contains all user behavior objects. Finally, normalizing the target matrix can obtain the cosine similarity matrix between commodities.

[0137] Among them, the similarity result matrix can be understood as a matrix used to represent the similarity results between the target behavior object and other behavior objects in at least two user behavior objects; for example, this similarity result matrix can be a cosine similarity matrix. Taking the commodity corresponding to the user behavior as the user behavior object as an example, in the process of calculating the similarity between commodities, after obtaining the adjusted co-occurrence matrix, add this co-occurrence matrix, and normalize the added matrix, then the cosine similarity matrix between commodities can be obtained.

[0138] This similarity result can be understood as data representing the similarity degree between two user behavior objects; for example, this similarity result can be a similarity, and this similarity can be any value within the interval [0, 1]. When the similarity is closer to 1, the two user behavior objects are more similar; when the similarity is closer to 0, the two user behavior objects are less similar.

[0139] Specifically, the data processing model training method provided in this specification can construct a user behavior object matrix based on at least two user behavior objects; then, by performing similarity calculation on the user behavior object matrix, a similarity result matrix corresponding to at least two user behavior objects is obtained; for example, performing similarity calculation on the user behavior object matrix can be understood as adding the user behavior object matrix and normalizing the obtained matrix to obtain a cosine similarity matrix. Another example is that performing similarity calculation on the user behavior object matrix can also be understood as adding the user behavior object matrix to obtain a target matrix, and calculating the Euclidean distance of the target matrix to obtain a Euclidean distance matrix (i.e., the similarity result matrix) containing the Euclidean distance calculation results. The Euclidean distance value of the calculation result represents the similarity between two user behavior objects; the larger the Euclidean distance value, the smaller the similarity, and the smaller the Euclidean distance value, the larger the similarity. Determine the similarity results between the target behavior object and other behavior objects among at least two user behavior objects from the similarity result matrix. Thus, it is possible to accurately determine the similar behavior objects that are relatively similar to each user behavior object based on the similarity results, which is convenient for subsequent sample expansion based on relatively similar similar behavior objects, and can improve the quantity of sequence-type sample features while improving the performance of sequence-type sample features.

[0140] Taking the application of the data processing model training method provided in this specification in the commodity recommendation scenario as an example, the data processing model training method is described. For sequence-type feature data, there may be situations such as fewer samples, and a replacement data augmentation method based on collaborative filtering is proposed to augment the original data. For the sake of explanation, assume that the dataset contains three sequence-type features: commodity ID, commodity category ID, and user behavior sequence, and these three sequence-type features form an initial sequence-type sample feature. First, create a correspondence table between user IDs and commodity IDs. Secondly, construct a relationship matrix between different commodities, and calculate the similarity between commodities through the cosine similarity calculation method to obtain a similarity result matrix, thereby determining the similarity results. Specifically, among them, the user behavior object can be the commodity corresponding to the user behavior; based on this, in the process of calculating the similarity between commodities, first, create a co-occurrence matrix of commodities. Then, for each user, add 1 to each pair of commodities in the list of commodities he interacts with in the co-occurrence matrix, and finally normalize the co-occurrence matrix to obtain the cosine similarity matrix between commodities. Among them, the list of commodities refers to the correspondence list between users and commodities.

[0141] It should be noted that in one or more embodiments provided in this specification, a relationship table between sample users and user behavior objects also needs to be constructed. For example, a relationship table between sample users and commodities is created. According to the interaction information between users and commodities in the original dataset (i.e., the initial sequence-type sample features), a corresponding list of user IDs and commodity IDs is created, and this corresponding list contains the commodity IDs that each user has interacted with. Among them, in one or more embodiments provided in this specification, the interaction between sample users and commodities refers to the cross-operation behaviors of sample users such as browsing commodities, purchasing commodities, and collecting commodities for the commodities.

[0142] Step 3: Determine the corresponding similar behavior object for the target behavior object based on the similarity result.

[0143] In one or more embodiments provided in this specification, the similarity result is the similarity degree;

[0144] The determining the corresponding similar behavior object for the target behavior object based on the similarity result includes:

[0145] Determine the target similarity degree that is greater than or equal to the preset similarity threshold from the similarity degrees between the target behavior object and the other behavior objects;

[0146] Determine the other behavior objects corresponding to the target similarity degree as at least two candidate similar behavior objects of the target behavior object;

[0147] Sort the at least two candidate similar behavior objects based on the target similarity degree to obtain a sequence of candidate similar behavior objects;

[0148] Select the preset first number of the candidate similar behavior objects from the sequence of candidate similar behavior objects from top to bottom as the similar behavior objects corresponding to the target behavior object.

[0149] Among them, the preset similarity threshold can be set according to the actual application scenario, and this specification does not make specific limitations on this; for example, in the case where the similarity degree is any value in the interval [0, 1], the preset similarity threshold can be the value 0.7. The candidate similar behavior objects can be understood as user behavior objects among the other behavior objects that are relatively similar to the target behavior object. The preset first number can be set according to the actual application scenario, and this specification does not make specific limitations on this. For example, the preset first number can be 10 or 5.

[0150] From the candidate similar behavior object sequence, select a preset first number of the candidate similar behavior objects from top to bottom as the similar behavior objects corresponding to the target behavior object. It can be understood that select the first 10 or the first 5 candidate similar behavior objects in the candidate similar behavior object sequence from top to bottom as the similar behavior objects corresponding to the target behavior object.

[0151] It should be noted that sorting at least two candidate similar behavior objects based on the target similarity can be understood as that the candidate similar behavior object with a larger target similarity is ranked at the forefront, and the candidate similar behavior object with a smaller target similarity is ranked at the last column.

[0152] Continuing with the above example, where the target behavior object is the target commodity and the other behavior object is another commodity. After calculating the similarity between commodities through cosine similarity calculation to obtain a similarity matrix, the similarity between the target commodity and other commodities can be determined. After obtaining the similarity, it is necessary to compare the similarity with a preset similarity threshold of 0.7 to determine a target similarity greater than or equal to the preset similarity threshold of 0.7, which indicates that the target commodity is relatively similar to other commodities. Based on the other commodities corresponding to the target similarity, construct a set of similar commodities for the target commodity; finally, sort using the target commodity similarity and take the top five other commodities as the set of similar commodities for the target commodity; thus accurately determining the similar behavior object for the target behavior object. Among them, the target commodity is any commodity included in the commodity ID sequence type feature. That is to say, for each commodity included in the commodity ID sequence type feature, a set of similar commodities is determined. The other commodities are other commodities except the target commodity in the commodity ID sequence type feature.

[0153] In one or more embodiments provided in this specification, if no set of similar commodities is obtained after collaborative filtering calculation, find the category to which the target commodity belongs, and select the top 20% of the commodities with the highest appearance frequency in this category as the set of similar items for the target commodity, so as to facilitate subsequent sample expansion based on similar behavior objects. Specifically, the similar result is the similarity;

[0154] Correspondingly, determining the corresponding similar behavior object for the target behavior object based on the similar result includes:

[0155] Determining the corresponding similar behavior object for the target behavior object based on the similar result includes:

[0156] In the case where the similarity between the target behavior object and the other behavior object is less than the preset similarity threshold, determine the object type information of the target behavior object;

[0157] From the at least two user behavior objects, determine at least two candidate similar behavior objects corresponding to the object type information, and determine the object quantity of each candidate similar behavior object;

[0158] Sort the at least two candidate similar behavior objects based on the object quantity to obtain a candidate similar behavior object sequence;

[0159] From the candidate similar behavior object sequence, select a preset second quantity of the candidate similar behavior objects from top to bottom as the similar behavior objects corresponding to the target behavior object.

[0160] Among them, the object type information can be understood as the information characterizing the type of the user behavior object; for example, taking the user behavior object as the commodity corresponding to the user behavior, the object type information is the commodity type of the commodity. For example, the commodity type can be facial cleanser type, coat type, marker type, etc.

[0161] The preset second quantity can be set according to the actual application scenario, and this specification does not make specific limitations. For example, the preset second quantity can be 20% or the top 30%.

[0162] Selecting a preset second quantity of the candidate similar behavior objects from top to bottom as the similar behavior objects corresponding to the target behavior object from the candidate similar behavior object sequence can be understood as selecting the top 20% of the candidate similar behavior objects in the candidate similar behavior object sequence from top to bottom as the similar behavior objects corresponding to the target behavior object.

[0163] Continuing with the above example, after obtaining the similarity matrix by calculating the similarity between commodities through the cosine similarity calculation method, the similarity between the target commodity and other commodities can be determined. After obtaining the similarity, it is necessary to compare the similarity with a preset similarity threshold of 0.7. In the case where the similarities are all less than the preset similarity threshold of 0.7, determine the commodity type of the target commodity, and determine the same-type commodities corresponding to this commodity type from other commodities. Use these same-type commodities as the candidate similar commodities of the target commodity. Then determine the number of times each commodity appears in the candidate similar commodities, that is, the commodity quantity of each candidate similar commodity in the candidate similar commodities. Sort the candidate similar commodities using this commodity quantity, and use the top 20% of the candidate similar commodities as the set of similar commodities of the target commodity; thus accurately determining the similar behavior objects for the target behavior object.

[0164] In one or more embodiments provided in this specification, the target behavior object may be the user behavior object with the earliest interaction time in the initial sequence-type sample features; that is to say, the target behavior object may be the user behavior object at the first position in the sequence in the initial sequence-type sample features. The interaction time of this target behavior object is earlier than the interaction times of other user behavior objects among at least two user behavior objects. For example, this target behavior object is the target commodity, and this target commodity is the commodity with the earliest time point in the time series-type features. That is to say, the target commodity ranked first in the time series-type features. Therefore, for the commodity ID that is the earliest in the time series-type features, determine a set of similar commodities similar to it.

[0165] In one or more embodiments provided in this specification, in the face of situations such as fewer samples in the data set, a method for augmenting the original data set based on self-supervised and other data set augmentation methods is proposed to improve the prediction effect of the model from the data itself level. During the process of sample data augmentation, after determining the similar behavior objects corresponding to the user behavior objects, it is necessary to perform sample augmentation on the user behavior objects and the similar behavior objects. The specific steps are as follows: The number of initial sequence-type sample features is at least two;

[0166] Performing sample augmentation based on the user behavior object and the similar behavior object to obtain augmented sequence-type sample features includes:

[0167] Determine the user behavior object information of each initial sequence-type sample feature among at least two initial sequence-type sample features;

[0168] Based on the user behavior object information, determine the sequence-type sample feature to be augmented from the at least two initial sequence-type sample features;

[0169] Based on the similar behavior object, process the user behavior object in the sequence-type sample feature to be augmented to obtain the augmented sequence-type sample features.

[0170] Among them, the user behavior object information may be the number of object of the user behavior object included in the initial sequence-type sample feature. Or the object size of multiple user behavior objects included in the initial sequence-type sample feature. Subsequently, the sequence-type sample feature to be augmented can be determined from at least two initial sequence-type sample features according to the number of objects or the object size.

[0171] Specifically, determine the number of user behavior objects for each of at least two initial sequential sample features; compare the number of user behavior objects with a preset object number threshold, and determine the initial sequential sample features with the number of user behavior objects greater than the preset object number threshold among the at least two initial sequential sample features as the sequential sample features to be augmented; process the user behavior objects in the sequential sample features to be augmented based on similar behavior objects to obtain augmented sequential sample features; thereby augment the original data set and improve the prediction effect of the model from the data itself level.

[0172] Continuing with the above example, where the user behavior object information refers to the number of user behavior objects. Based on this, first, it is necessary to determine the number of product Ids included in each sequential behavior feature, and determine the sequential behavior features of product Ids with the number of product Ids greater than 10 as the sequential behavior features to be augmented. It should be noted that the number of product Ids also represents the length of the sequential feature of product Ids. When the number of product Ids is 10, the length of the sequential feature of product Ids is also 10. Secondly, from the sequential features of product Ids with a length greater than 10, determine the product ID (target product) with the earliest appearance time; and determine the set of similar products corresponding to this product ID. Augment the sequential feature of product Ids based on this set of similar products to obtain augmented sequential sample features.

[0173] In one or more embodiments provided in this specification, the processing of the user behavior objects in the sequential sample features to be augmented based on the similar behavior objects to obtain augmented sequential sample features includes:

[0174] Determine the behavior object to be augmented and the corresponding object type information of the behavior object to be augmented from at least two user behavior objects included in the sequential sample features to be augmented;

[0175] Select a target similar behavior object from the similar behavior objects corresponding to the behavior object to be augmented, and determine the target object type information of the target similar behavior object;

[0176] Replace the behavior object to be augmented in the sequential sample features to be augmented with the target similar behavior object, and replace the object type information corresponding to the behavior object to be augmented in the sequential sample features to be augmented with the target object type information to obtain the augmented sequential sample features.

[0177] Among them, the behavior object to be augmented can be understood as the user behavior object that needs to be augmented; the object type information of the behavior object to be augmented can be understood as the object type information of the behavior object to be augmented. The target object type information can be understood as the object type information of the target similar behavior object.

[0178] Specifically, a behavior object to be augmented is determined from at least two user behavior objects included in the sample feature of the sequence type to be augmented. The behavior object to be augmented can be understood as the user behavior object at the first sequence position in the sample feature of the sequence type to be augmented, or the behavior object to be augmented can be understood as the user behavior object with the earliest generation time in the sample feature of the sequence type to be augmented.

[0179] After determining the behavior object to be augmented, determine the type information of the object to be augmented corresponding to the behavior object to be augmented; then select a target similar behavior object from the similar behavior objects corresponding to the behavior object to be augmented, and determine the type information of the target object of the target similar behavior object; the target similar behavior object can be any one of the similar behavior objects. Alternatively, the target similar behavior object can be a similar behavior object with a relatively high similarity to the behavior object to be augmented among the similar behavior objects.

[0180] Replace the behavior object to be augmented in the sample feature of the sequence type to be augmented with the target similar behavior object, and replace the type information of the object to be augmented corresponding to the behavior object to be augmented in the sample feature of the sequence type to be augmented with the type information of the target object, to obtain an augmented sample feature of the sequence type; thereby augment the original data set and improve the prediction effect of the model from the data itself level.

[0181] Continuing with the above example, Figure 6 is a schematic diagram of data augmentation in a data processing model training method provided by an embodiment of this specification; see Figure 6 , where the data 1 included in the original data set can be understood as the initial sample feature of the sequence type, and the original data set can include multiple pieces of data. The initial sample feature of the sequence type includes a commodity ID feature (i.e., a commodity ID time series feature), a category ID feature (a category ID time series feature), and a user behavior sequence feature. Based on this, after determining the set of similar commodities, randomly select a similar commodity ID from the set of similar commodities, and use the similar commodity ID to replace the target commodity in the commodity ID time series feature. Then determine the category ID time series feature and the user behavior sequence feature corresponding to the commodity ID time series feature. For the category ID time series feature of the commodity, replace the category ID of the target commodity with the category ID corresponding to the similar commodity; the user behavior sequence feature remains unchanged, and thus a new sample feature of the sequence type is constructed based on an original user behavior record (i.e., Figure 6 the data 2 in the data set after data augmentation in

[0182] Step 406: Perform feature enhancement on the initial sequential sample features to obtain the sequential sample features, and perform feature enhancement on the initial non-sequential sample features to obtain the non-sequential sample features.

[0183] In one or more embodiments provided in this specification, the data processing model training method in this specification proposes a full-feature contrast learning model, which takes into account both sequential data features and non-sequential data features, innovates the data augmentation method of the existing model, and also optimizes the existing model from the perspectives of training time and training cost. While exceeding the prediction effect of the existing model, it also meets the requirements of the actual application scenario. Specifically, the performing feature enhancement on the initial sequential sample features to obtain the sequential sample features includes:

[0184] Input the initial sequential sample features into the first data processing module in the data processing model to be trained;

[0185] Use the feature processing unit in the first data processing module to process the initial sequential sample features to obtain the sequential sample features to be enhanced;

[0186] Determine the target data augmentation rule from at least two data augmentation rules, and use the target data augmentation rule to perform data augmentation on the sequential sample features to be enhanced to obtain the sequential sample features.

[0187] Among them, the first data processing module can be understood as the module in the data processing model to be trained for processing sequential sample features; the first data processing module can be a data processing model. For example, the first data processing module can be a sequential contrast learning model. Or the first data processing module can be one or more network layers for processing sequential sample features.

[0188] The feature processing unit can be understood as the input layer or embedding layer integrated in the first data processing module for processing sequential sample features. The sequential sample features to be enhanced can be understood as the features obtained by inputting the sequential sample features to be enhanced into the feature processing unit.

[0189] The data augmentation rule can be set according to the actual application scenario, and this specification does not make specific restrictions on this. For example, the data augmentation rule can be a method for vector augmentation (i.e., feature augmentation). For example, the vector augmentation operations include vector masking operation, vector cropping operation, and vector reordering operation. Among them, the vector masking operation sets a learnable tensor, which can be used to perform masking operations on feature vectors (i.e., initial sequence sample features, augmented sequence sample features); the vector cropping operation can perform random cropping operations on feature vectors; the vector reordering operation can perform reordering operations on feature vectors. In one or more embodiments provided in this specification, the data augmentation operation can be implemented through a data augmentation module in the data processing model. The data augmentation module can be understood as one or more network layers in the data processing model; or, the data augmentation module can be understood as a sub-model in the data processing model for performing data augmentation. Based on this data processing module, the three vector augmentation schemes provided in this specification perform data augmentation after the original data passes through the embedding layer, so as to maintain the internal association between users and commodities in historical behaviors and improve the performance of the augmented sample data.

[0190] The target data augmentation rule can be understood as the data augmentation rule among at least two data augmentation rules for performing data augmentation on the sequence sample features to be augmented. The target data augmentation rule can be set according to the actual application scenario. Among them, the target data augmentation rule can be one or at least two.

[0191] Continuing with the above example, Figure 7 is a schematic diagram of training a sequence contrast learning model in a data processing model training method provided by an embodiment of this specification. Refer to Figure 7 , Figure 7 The records in can be understood as initial sequence sample features. For the initial sequence sample features, the sequence contrast learning model splices the sequence features to obtain the spliced sequence features; inputs the spliced sequence features into the input layer integrated by EasyRec (i.e., the sequence contrast learning model), and obtains the corresponding feature vector representation (i.e., the sequence sample features to be augmented) through the input layer integrated by EasyRec; finally, performs vector augmentation operations on the feature vector to obtain the positive sample pairs in contrast learning (i.e., the above-mentioned sequence sample features), thereby innovating the data augmentation method of the existing model, and also optimizing the existing model from the perspectives of training time and training cost. While exceeding the prediction effect of the existing model, it also meets the requirements of the actual application scenario.

[0192] Based on Figure 7It can be seen that subsequently, all the sequence-type sample features obtained after feature enhancement are input into the encoder in the sequence-type contrastive learning model. According to the results output by the encoder, calculations are performed through a loss function to obtain the loss value of the sequence-type contrastive learning function.

[0193] It should be noted that Figure 7 the encoder in the sequence-type contrastive learning model shares parameters with the encoders in other models (the encoders in the multi-layer perceptron model). It can be understood that this is to obtain the vector representation of the sequence features; different task encoders are different. The encoder is mainly to obtain the vector representation of the features. Figure 7 The shared parameter here is that the right side is for the contrastive learning task, and the left side can be the CTR task (optional). To generate the vector representation of the sequence features, the same encoder (such as all using a transformer) is required. The shared parameter is to ensure that the vector representations generated by the sequence features on both sides are the same.

[0194] In one or more embodiments provided in this specification, a full-feature contrastive learning model is proposed, which can optimize the data augmentation method of the existing model and further optimize the existing model from the perspectives of training time and training cost, so as to exceed the prediction effect of the existing model. Specifically, the method of determining the target data augmentation rule from at least two data augmentation rules and using the target data augmentation rule to perform data augmentation on the sequence-type sample features to be augmented to obtain the sequence-type sample features includes:

[0195] Select two target data augmentation rules from the preset first data augmentation rule, the preset second data augmentation rule, and the preset third data augmentation rule;

[0196] Use the two target data augmentation rules to perform data augmentation on the sequence-type sample features to be augmented to obtain the first enhanced sequence-type sample features and the second enhanced sequence-type sample features;

[0197] Take the first enhanced sequence-type sample features and the second enhanced sequence-type sample features as the sequence-type sample features.

[0198] Among them, the preset first data augmentation rule, the preset second data augmentation rule, and the preset third data augmentation rule can be set according to the actual application scenario, and this specification does not make specific restrictions on this. For example, the first data augmentation rule can be a vector masking operation, the preset second data augmentation rule can be a vector clipping operation, and the preset third data augmentation rule can be a vector reordering operation. Based on this, selecting two data augmentation rules can be understood as randomly selecting two data augmentation rules from the preset first data augmentation rule, the preset second data augmentation rule, and the preset third data augmentation rule. Moreover, randomly selecting two data augmentation rules can be to select two identical data augmentation rules. For example, randomly select to use the preset first data augmentation rule to perform data augmentation on the target sequence-type behavior vector twice.

[0199] Continuing with the above example, Figure 8 is a schematic diagram of sequence feature data augmentation in a data processing model training method provided by an embodiment of this specification. Based on Figure 8 This specification details three data augmentation schemes provided by the data processing model training method: vector masking, vector clipping, and vector reordering.

[0200] Among them, vector masking can be understood as item mask, which is a technology based on the BERT model applied in natural language processing and is called zero-masking in the BERT model, used to extract key information from text and perform tasks such as text classification and named entity recognition. This item masking technology can solve problems such as missing context information, ambiguity, and context in the text. The first data processing module (such as the CL4SRec model) in the data processing model training method provided by this specification randomly masks a certain proportion of items (for example, replaces them with 0) in each user's historical sequence, but this will, to a certain extent, damage the inherent attributes of the items.

[0201] Based on this, to avoid the problem that the item masking operation will damage the inherent attributes of the items, the data processing model training method provided by this specification makes improvements. The specific improvement steps are as follows:

[0202] First, for each user's historical behavior sequence s u , which contains I = (i1, i2,..., i n ) n items (l represents the feature set at a certain moment in the user's historical record, with a size of m. In this dataset, l represents three features: commodity ID, commodity category ID, and behavior type). Each user's historical behavior sequence s u is passed through the embedding layer to obtain V = (v1, υ2,..., v n), the dimension of υ is n*E, where E is the embedding size, and its calculation formula is shown in the following formula (1):

[0203]

[0204] Secondly, randomly mask a certain proportion of items aug mask =(index1, index2,..., index mask ), where L mask =p mask *n, and L mask refers to the length of the sequence after masking; p mask refers to the ratio of masking the sequence length; aug mask represents the indices of the high-dimensional vectors selected to be masked in the sequence. index refers to the high-dimensional vector index, and this vector will be replaced by a learnable vector [υector m . In summary, the calculation formulas of the item masking algorithm are shown in formulas (2) and (3):

[0205]

[0206]

[0207] Among them, refers to the user's historical behavior sequence s u .

[0208] Since the types of products browsed by users are likely to be stable within a certain period of time. For example, if a user wants to buy a facial cleanser, he is likely to click on many facial cleanser brands or different merchants to decide which one to buy. It can be seen that the historical interactions between users and products are likely to be similar within a certain period of time. Therefore, through this item masking algorithm, two different views obtained from the historical sequence by the same user through this algorithm can still retain the user's purchase intention, and the learnable variables can enable the model to learn and discover more internal associations between users and products.

[0209] Among them, vector cropping can be understood as item crop, which is a technique widely used in various image recognition and analysis tasks in computer vision. Its purpose is to crop out the unnecessary regions in the image, so as to concentrate on processing the important regions and improve the efficiency and accuracy of image processing. In the first data processing module (such as the CL4SRec model) in the data processing model training method provided in this specification, the original data of each user is randomly cropped; in the data processing model training method provided in this specification, specifically, the original data is first passed through the embedding layer and then randomly cropped to complete data augmentation. The specific execution steps are as follows:

[0210] First, for each user's historical behavior sequence s u After passing through the embedding layer, we get V=(υ1, υ2,..., υ n ), which has the same meaning as the parameters described in the item masking operation and will not be elaborated here.

[0211] Second, randomly crop a certain proportion of items aug crop =(index1, index2,..., index crop ), where L crop =p crop *n, L crop refers to the length of the sequence after cropping, p crop refers to the ratio of the reserved sequence length; aug crop represents the indices of the high-dimensional vectors selected to be cropped in the sequence. Finally, just delete the items corresponding to the indices included in aug crop to obtain the final data-augmented view. In summary, the calculation formula of the item cropping algorithm is shown in the following formula (4).

[0212]

[0213] Among them, refers to the user's historical behavior sequence s u after vector cropping.

[0214] The data augmentation effect achieved by this item cropping calculation can be reflected in two aspects. On the one hand, it provides a local view of the user's historical sequence. Without a comprehensive analysis of the user, the user representation model is enhanced by learning the user's general preferences. On the other hand, in the contrast learning algorithm, if the views obtained after two croppings have no intersection, they can be regarded as a prediction task for future time points, that is, let the model predict the change in the user's purchase preference in the future.

[0215] Among them, vector reordering can be understood as item reorder. Sequence recommendation tasks often predict data at the next time point based on the passage of time, believing that the items interacting with the user are order-related. However, due to complex external environmental factors, the interaction order between the user and the items can also be flexible. Therefore, item reordering is used for data augmentation.

[0216] Based on this, for the initial sequential sample features, after inputting them into the sequential contrast learning model. First, the sequential contrast learning model splices the initial sequential sample features through the embedding layer to obtain the spliced initial sequential sample features; the spliced sequential features are input into the input layer of the model integration to obtain the corresponding feature vectors (i.e., the sequential sample features to be augmented); finally, two random data augmentation operations are performed on the feature vectors. The data augmentation operation can be the above-mentioned vector masking, vector clipping, or vector reordering. Both of the two data augmentation operations are randomly selected from the above three data augmentation methods for data augmentation. After data augmentation, two augmented sequential sample features corresponding to the two data augmentation operations are obtained (i.e., the positive sample pairs in contrast learning). Thus, the data augmentation method of the existing model is optimized, and the existing model is further optimized from the perspectives of training time and training cost, exceeding the prediction effect of the existing model.

[0217] In one or more embodiments provided in this specification, the data processing model training method of this specification proposes a full-feature contrast learning model, which considers both sequential data features and non-sequential data features, and innovates the data augmentation method of the existing model. By performing feature augmentation on the non-sequential data features, the existing model is optimized from the perspectives of training time and training cost, while improving the model prediction effect and meeting the requirements of the actual application scenario. The specific method is as follows. The feature augmentation of the initial non-sequential sample features to obtain the non-sequential sample features includes:

[0218] Input the initial non-sequential sample features into the second data processing module in the data processing model to be trained;

[0219] Perform feature masking on the initial non-sequential sample features to obtain enhanced non-sequential sample features;

[0220] Use the feature processing unit in the second data processing module to process the enhanced non-sequential sample features to obtain the non-sequential sample features.

[0221] Among them, the second data processing module can be understood as the module in the data processing model to be trained for processing non-sequential sample features; the second data processing module can be one or more data processing models. For example, the second data processing module can be a sequential contrast learning model and / or a multi-layer perceptron model. Or the second data processing module can be one or more network layers for processing non-sequential sample features.

[0222] In the data processing model training method provided in this specification, feature enhancement can be performed on the initial non-sequential sample features by means of feature masking. In one or more embodiments provided in this specification, in the data processing model training method provided in this specification, two data enhancement methods for the initial non-sequential sample features will be introduced. The data enhancement method can be a feature masking (or called feature masking, vector masking) method. Since non-sequential sample features generally do not depend on the relationship of time order, commodity reordering is no longer applicable, and commodity cropping is similar to feature masking. Therefore, the data masking is improved here. Two feature masking methods are provided in the data processing model training method provided in this specification, and any one of them will be determined as the data enhancement method later. The specific two feature masking methods are as Figure 9 shown Figure 9 is a schematic diagram of two feature masking methods in a data processing model training method provided in an embodiment of this specification; specifically, Figure 9 The feature masking method 1 on the left is the first feature masking method.

[0223] For the first feature masking method, each initial non-sequential sample feature is regarded as a whole and directly masked. Among them, the masking method is to directly set the feature to be masked to a zero vector. Specifically, assume that the non-sequential feature at the current moment is F = (f1, f2,..., f n ) with a total of n features. First, after passing through the data input layer (embedding layer), we get V = (υ1, υ2,..., υ n ), and the dimension of υ is n*E, where E is the size of the embedding layer. Secondly, randomly mask a certain proportion of items aug mask = (index1, index2,..., index mask ), where L mask = p mask *n, and aug mask represents the indexes selected to be masked in the sequence, and this vector will be replaced with a zero vector of size 1*E. In summary, the calculation formula of the first feature masking method is shown in the following formulas (5) and (6):

[0224]

[0225]

[0226] Among them, refers to the non-sequential features after feature masking using the first feature masking method.

[0227] Figure 9 The feature masking method 2 on the right is the second feature masking method. This second feature masking method is similar to the model-level data augmentation method mentioned in the Duorec model. It mainly randomly masks the feature vector representation after passing through the data input layer. Figure 9 The square framed by the dashed box in is the masked information. The specific approach is to multiply the original feature matrix by a Bernoulli random variable matrix to randomly mask some information of the features with a probability of p. In summary, the calculation formula of the second feature masking method is shown in the following formulas (7) and (8):

[0228]

[0229] B ~ Bernoulli(p) Formula (8).

[0230] Among them, refers to the non-sequential features after feature masking using the second feature masking method. B ~ Bernoulli(p) refers to the Bernoulli random variable matrix.

[0231] The feature processing unit can be understood as the input layer or embedding layer integrated in the second data processing module, and is used to process the enhanced non-sequential sample features. This encoding unit can be understood as the encoder in the non-sequential data processing module, and this encoding unit can adopt MaskNet.

[0232] It should be noted that the Transformer encoder usually has good results when dealing with time series-related features, but its training time is long and it is not very applicable when the number of features is large. Therefore, the data processing model training method provided in this specification uses MaskNet as the feature representation module (i.e., encoder) of the non-sequence data processing module (i.e., non-sequence contrast learning model). Compared with the Transformer, MaskNet has better performance in encoding large-scale features. Specifically, MaskNet proposes an instance-guided mask scheme that uses element-wise product in both the feature embedding layer and the feed-forward layer of the DNN. This instance-guided masking method can dynamically incorporate global context information into the feature embedding layer and the feed-forward layer, highlighting important features when the dataset contains a large number of types and quantities. This shows that the model has good performance in actual application scenarios. Therefore, the data processing model training method provided in this specification selects it as the encoder to form the feature representation module.

[0233] Continuing with the above example, Figure 10 is a schematic diagram of the training of the non-sequence contrast learning model in a data processing model training method provided by an embodiment of this specification. Refer to Figure 10 , for the non-sequence features in the dataset (i.e., initial non-sequence sample features), input them into the non-sequence contrast learning model. First, perform data augmentation operations on the non-sequence features. The data augmentation operation selects the feature masking method, that is, randomly mask several non-sequence features as a whole with the feature. Secondly, the enhanced non-sequence features pass through the input layer integrated by EasyRec to obtain the corresponding feature vector representation after data augmentation (i.e., non-sequence sample features).

[0234] Based on Figure 10 it can be seen that after obtaining the non-sequence sample features, they will be input into the encoder (MaskNet can be selected as the encoder because this model can highlight important features and has a faster training speed). Finally, according to the results output by the encoder, calculate the loss value using the loss function. Among them, L2_loss can be selected as the loss function.

[0235] It should be noted that the Figure 10 other models included in this can be understood as the first data processing module and the data prediction module in one or more embodiments provided in this specification; for the Figure 10 explanation of other models in this, refer to the description of the first data processing module and the data prediction module in one or more embodiments provided in this specification.

[0236] Step 408: Based on the sequential sample features and the non-sequential sample features, train the data processing model to be trained to obtain a data processing model.

[0237] In one or more embodiments provided in this specification, in the data processing model training method provided in this specification, related work on the application of contrastive learning in the recommendation scenario usually considers sequential sample features and non-sequential sample features separately, while the full-feature contrastive learning model provided in this specification considers both types of features at the same time. By improving and optimizing the data augmentation method of the CL4SRec model and using it as part of the sequential contrastive learning model, the performance of the full-feature contrastive learning model is provided.

[0238] In one or more embodiments provided in this specification, the training of the data processing model to be trained based on the sequential sample features and the non-sequential sample features to obtain the data processing model includes:

[0239] Use the first data processing module in the data processing model to be trained to process the sequential sample features to obtain a first sample processing result;

[0240] Use the second data processing module in the data processing model to be trained to process the non-sequential sample features to obtain a second sample processing result;

[0241] Train the first data processing module based on the first sample processing result and train the second data processing module based on the second sample processing result until the model training stop condition is reached to obtain the data processing model.

[0242] Among them, the first sample processing result can be understood as the loss value for training the first data processing module. The second sample processing result can be understood as the loss value for training the second data processing module.

[0243] Specifically, in the data processing model training method provided in this specification, after enhancing the initial sequential sample features and initial non-sequential sample features to obtain sequential sample features and non-sequential sample features, for the sequential sample features, they can be input into the encoding unit in the first data processing module for encoding processing, so as to obtain the first sample processing result determined based on the sequential sample data. For the non-sequential sample features, they can be input into the encoding unit in the second data processing module for encoding processing, so as to obtain the second sample processing result determined based on the non-sequential sample data. Then, based on the first sample processing result, the parameters of the first data processing module are adjusted, and based on the second sample processing result, the parameters of the second data processing module are adjusted until the model training stop condition is reached, and a trained data processing model is obtained. Thus, during the model training process, the influence of sample data on model training is fully considered, avoiding the problem of poor model training efficiency caused by sample data, and meeting the needs of practical applications.

[0244] Continuing with the above example, after obtaining the enhanced sequential sample features, they will be input into the encoder in the sequential contrastive learning model. All the enhanced sequential sample features are input into the encoder in the sequential contrastive learning model, and according to the results output by the encoder, through calculation using the loss function, the loss value of the sequential contrastive learning function is obtained. After obtaining the enhanced non-sequential sample features, they will be input into the encoder in the non-sequential contrastive learning model. Then, according to the results output by the encoder, the loss value of the non-sequential contrastive learning function is calculated using the loss function. After obtaining the loss value of the sequential contrastive learning function and the loss value of the non-sequential contrastive learning function, the parameters of the sequential contrastive learning model are adjusted based on the loss value of the sequential contrastive learning function; the parameters of the non-sequential contrastive learning model are adjusted based on the loss value of the non-sequential contrastive learning function until the model training stop condition is reached, and a trained full-feature contrastive learning model is obtained.

[0245] In one or more embodiments provided in this specification, in the data processing model training method provided in this specification, in the related work on the application of contrastive learning in the recommendation scenario, sequential features and non-sequential features are usually considered separately. However, the full-feature contrastive learning model provided in this specification considers both types of features at the same time; by optimizing the CL4CTR model in terms of training speed and training cost, the encoder is replaced with MaskNet so that it can be applied in real application scenarios and serves as part of the non-sequential contrastive learning model. Finally, the sequential contrastive learning model and the non-sequential contrastive learning model are combined as the full-feature contrastive learning model to provide assistance for predicting in actual recommendation application scenarios. Specifically, the second data processing module includes a non-sequential data processing module and a data prediction module;

[0246] Processing the non-sequential sample features by using the second data processing module in the to-be-trained data processing model to obtain a second sample processing result, including:

[0247] Encoding the non-sequential sample features by using the encoding unit in the non-sequential data processing module to obtain the non-sequential data processing result;

[0248] Processing the non-sequential sample features by using the data prediction module to obtain a data prediction result;

[0249] Taking the non-sequential data processing result and the data prediction result as the second sample processing result.

[0250] Among them, the non-sequential data processing module can be understood as a non-sequential contrastive learning model in the data processing model. The data prediction module can be understood as a multi-layer perceptron model in the data processing model. The data processing model can be a full-feature contrastive learning model. The non-sequential data processing result can be understood as the loss value for training the non-sequential data processing module. For example, it can be the loss value of the non-sequential contrastive learning function. The data prediction result can be understood as the loss value for training the data prediction module. For example, it can be the loss value of the multi-layer perceptron model.

[0251] The encoding unit can be understood as an encoder in the non-sequential data processing module, and the encoding unit can adopt a transformer.

[0252] Continuing with the above example, after using the non-sequential contrastive learning model to perform feature enhancement to obtain the enhanced non-sequential sample features, they will be input into the encoder in the non-sequential contrastive learning model (MaskNet can be selected as the encoder because this model can highlight important features and has a faster training speed). Then, according to the result output by the encoder, the loss value of the non-sequential contrastive learning function is calculated by using a loss function, and L2_loss can be selected as the loss function. In addition, the non-sequential sample features are input into the multi-layer perceptron model, and the loss value of the multi-layer perceptron model is calculated according to the result output by the multi-layer perceptron model. The loss value of the non-sequential contrastive learning function and the loss value of the multi-layer perceptron model are used as the second sample processing result for model training. Subsequently, the parameters of the non-sequential contrastive learning model can be adjusted based on the loss value of the non-sequential contrastive learning function; the parameters of the multi-layer perceptron model can be adjusted based on the loss value of the multi-layer perceptron model until the model training stop condition is reached.

[0253] In the embodiments provided in this specification, the non-sequential sample features are input into the non-sequential data processing module, the initial non-sequential sample features are data-augmented, and the obtained augmented non-sequential sample features are input into the feature processing unit in the non-sequential data processing module to obtain non-sequential sample features; finally, the non-sequential sample features are input into the encoding unit in the non-sequential data processing module for encoding processing to obtain a non-sequential data processing result. Moreover, the data prediction module is used to process the non-sequential sample features to obtain a data prediction result. By using the non-sequential data processing result and the data prediction result as the second sample processing result, it is convenient to perform model training based on this second sample processing result, further optimizing the model performance metrics of the current application scenario and meeting the needs of practical applications.

[0254] In one embodiment provided in this specification, the use of the data prediction module to process the non-sequential sample features to obtain a data prediction result includes:

[0255] Input the non-sequential sample features into the data prediction module, and use the feature processing unit in the data prediction module to process the non-sequential sample features to obtain processed non-sequential sample features;

[0256] Input the processed non-sequential sample features into the encoding unit in the data prediction module for encoding processing to obtain the data prediction result.

[0257] Among them, the feature processing unit can be understood as the input layer or embedding layer integrated in the data prediction module. This encoding unit can be understood as the encoder in the data prediction module, and this encoding unit can adopt a multi-layer perceptron (MLP).

[0258] Following the above example, input the non-sequential sample features into the embedding layer in the multi-layer perceptron model to obtain the corresponding feature vector representation (i.e., the processed non-sequential sample features), and then input this feature vector representation into the multi-layer perceptron for binary classification prediction (which can be click-through rate prediction); calculate the multi-layer perceptron loss value based on the prediction result, so as to predict the click-through rate, which is convenient for improving the performance of the model based on this multi-layer perceptron loss value in the future.

[0259] In one or more embodiments provided in this specification, the loss function module of this model mainly consists of loss_ctr (i.e., the data prediction result) calculated by the basic model (i.e., the multi-layer perceptron model), cl_loss seq_model (i.e., the first sample processing result) calculated by the sequential contrast learning model and cl-loss none_seq_model(i.e., non-sequential data processing results). Among them, loss_ctr is a binary classification task, and the cl-loss calculated by the contrast model in this model none_seq_model is different from the calculation method of the contrast learning function loss in the sequential contrast learning model. Because in actual application scenarios, the number of non-sequential features is relatively large, such as thousands or more. Converting them into low-dimensional vector representations will result in a large vector matrix. If the calculation method in the sequential contrast learning model is adopted, it will lead to an excessive computational load of the model and a too high resource occupancy. Compared with the calculation method of the contrast learning function loss in the sequential contrast learning model, the calculation of cl-loss in the contrast model is simplified none_seq_model process (the feature alignment loss is included in the cl-loss none_seq_model ).

[0260] Continuing with the above example, after obtaining the above three loss values (i.e., the model prediction results), the performance of this model in the application scenario can be observed according to the model prediction results. The model is trained based on this loss value until the model training stop condition is reached. This model training stop condition can be set according to the actual application scenario. For example, the loss value reaches convergence

[0261] The data processing model training method provided in the embodiments of this specification, after determining the sequential samples and non-sequential samples of the data processing model to be trained, extracts features from the sequential samples and non-sequential samples in the data processing model to be trained, and performs feature enhancement on the initial sequential sample features and initial non-sequential sample features obtained by feature extraction to obtain sequential sample features and non-sequential sample features for model training; then trains the data processing model to be trained based on the sequential sample features and non-sequential sample features to obtain a data processing model. Thus, the influence of sample data on model training is fully considered during the model training process, avoiding the problem of poor model training efficiency caused by sample data and meeting the needs of actual applications

[0262] The following combines the attached Figure 11 , taking the application of the data processing model training method provided in this specification in the commodity recommendation scenario as an example, to further illustrate the data processing model training method. Among them Figure 11 shows the processing procedure flowchart of a data processing model training method provided in an embodiment of this specification, specifically including the following steps

[0263] Step 1102: Receive the dataset provided by the customer, and extract features from the data in the dataset to obtain data features

[0264] Step 1104: According to the data feature type, the data features are divided into sequential features (i.e., the sequential sample features in the above embodiments) and non-sequential features (i.e., the non-sequential sample features in the above embodiments).

[0265] Among them, the sequential feature refers to: the interaction records between users and commodities. For example, the interaction records can be composed of three time series features formed by commodity ID, commodity category ID, and user behavior sequence, where different time series features are in one-to-one correspondence within the same time period.

[0266] The non-sequential feature refers to: user-side data such as user age and user gender, and commodity-side features such as commodity price and commodity capacity. That is to say, all other features in the data features except the sequential features are non-sequential features.

[0267] Step 1106: Based on the replacement data augmentation method of collaborative filtering, the original sequential features are augmented.

[0268] The specific steps are as follows:

[0269] 1. Create a relationship table between users and commodities. According to the interaction information between users and commodities in the original dataset, create a correspondence table of user ID and commodity ID, which contains the commodity IDs that each user has interacted with.

[0270] 2. Calculate the similarity between items. First, create a co-occurrence matrix of items. Second, for each user, add 1 to each pair of items in the list of items he / she has interacted with in the co-occurrence matrix. Finally, normalize the co-occurrence matrix to obtain the cosine similarity matrix between items.

[0271] Step 1108: Determine the similarity between the target commodity and other commodities, and take the top five commodities in terms of similarity as the set of similar commodities of the target commodity.

[0272] Among them, the target commodity is the commodity with the longest time point in the time series. That is to say, for the commodity ID with the longest time in the time series, determine a set of similar commodities.

[0273] It should be noted that if no set of similar commodities is obtained after collaborative filtering calculation, find the category to which the target commodity belongs, and select the top 20% of the commodities with the highest appearance frequency in this category as the set of similar items of the target commodity.

[0274] Step 1110: Construct a new sequential feature data.

[0275] The specific steps are as follows:

[0276] 1. Determine the commodity ID time series feature with a length greater than 10.

[0277] 2. Determine the product ID (target product) with the longest appearance time from the time series features of product IDs with a length greater than 10; and determine the set of similar products corresponding to this product ID.

[0278] 3. Randomly select a product ID from the set of similar products and replace the target product.

[0279] 4. Determine the category ID and the time features of the behavior sequence corresponding to the time series features of this product ID. For the time series features of product category IDs, replace the category ID of the target product with the category ID corresponding to the similar product; the time series features of the behavior sequence remain unchanged, and thus construct a new piece of data based on an original user behavior record.

[0280] Step 1112: For sequential features, input them into a sequential contrastive learning model to obtain a loss value.

[0281] The specific steps are as follows:

[0282] 1. Concatenate a product ID, a product category ID, and a user behavior sequence to obtain the concatenated sequential feature.

[0283] 2. Input the concatenated sequential feature into the embedding layer in the model to obtain the corresponding feature vector representation.

[0284] 3. Perform two random vector augmentation operations on this feature vector representation to obtain positive sample pairs in contrastive learning, and this positive sample pair is two augmented sequential features.

[0285] Among them, the vector augmentation operations include vector masking, vector clipping, and vector reordering. The two random operations can be a data augmentation operation method.

[0286] 4. After obtaining two augmented sequential features, input the two augmented sequential features into the corresponding encoders respectively, and calculate the loss value according to the results output by the encoders.

[0287] Step 1114: For non-sequential features, input them into a non-sequential contrastive learning model to obtain a loss value.

[0288] The specific steps are as follows:

[0289] 1. Perform a data augmentation operation on the non-sequential features to obtain the augmented non-sequential features.

[0290] Among them, this data augmentation method selects the feature masking method.

[0291] 2. Input the augmented non-sequential features into the embedding layer in the model to obtain the corresponding feature vector representation after data augmentation.

[0292] 3. Input the feature vector representation after data augmentation into the encoder, and calculate the loss value according to the result output by the encoder.

[0293] Step 1116: Input the unsequenced features into the embedding layer to obtain the corresponding feature vector representation, and input the feature vector representation into a multi-layer perceptron for binary classification prediction; calculate the loss value according to the prediction result.

[0294] Step 1118: After completing the processes of the above steps 1112 to 1116, the model can be trained according to the model prediction results (the above three loss values) until the model training stop condition is reached.

[0295] Based on this, the data processing model training method provided in this specification can take into account the situation of missing or insufficient data samples in the real application scenario, and proposes a self-supervised dataset augmentation method based on the collaborative filtering algorithm to augment the original dataset. By augmenting the original dataset from the perspective of sample data to optimize the prediction effect of the model, compared with the related work that focuses on optimizing the model, this method has obvious advantages in terms of training cost and speed. Moreover, the ability of self-supervised and contrastive learning to effectively handle the sparsity, long-tail problems, etc. existing in the recommendation scenario data, combined with the actual application requirements, a contrastive learning model that simultaneously considers sequential features and unsequenced features is built to optimize the current model performance indicators.

[0296] See Figure 12 , Figure 12 shows a flowchart of another data processing method provided according to an embodiment of this specification. This data processing method is applied to the cloud and specifically includes the following steps.

[0297] Step 1202: Receive the user behavior data of the user sent by the terminal, where the user behavior data includes sequential behavior data and / or unsequenced behavior data;

[0298] Step 1204: Input the sequential behavior data and / or the unsequenced behavior data into the data processing model to obtain the data processing result corresponding to the user, where the data processing model is trained by sequential sample features and unsequenced sample features. The sequential sample features and the unsequenced sample features are obtained by extracting features from sequential samples and unsequenced samples in the data processing model to obtain initial sequential sample features and initial unsequenced sample features, and performing feature enhancement on the initial sequential sample features and the initial unsequenced sample features;

[0299] Step 1206: Send the data processing result to the terminal.

[0300] In one or more embodiments provided in this specification, the data processing model in another data processing method is obtained by training using sequential sample features and non-sequential sample features. The sequential sample features and non-sequential sample features are obtained by enhancing the initial sequential sample features and initial non-sequential sample features obtained by extracting features from sequential samples and non-sequential samples in the data processing model. Thus, during the model training process, the influence of sample data on model training is fully considered, avoiding the problem of poor model training efficiency caused by sample data. Moreover, the trained data processing model processes the sequential behavior data and / or non-sequential behavior data included in the user behavior data sent by the terminal, obtains a data processing result, and sends the data processing result to the terminal, thereby meeting the needs of practical applications through the data processing model.

[0301] The above is a schematic solution of another data processing method of this embodiment. It should be noted that the technical solution of another data processing method and the technical solution of the above-mentioned one data processing method belong to the same concept. For the details not described in detail in the technical solution of another data processing method, reference can be made to the description of the technical solution of the above-mentioned one data processing method.

[0302] See Figure 13 , Figure 13 shows a flowchart of an object recommendation model training method provided according to an embodiment of this specification, which specifically includes the following steps.

[0303] Step 1302: Determine the sequential samples and non-sequential samples of the object recommendation model to be trained, where the sequential samples and the non-sequential samples are user behavior data of sample users for sample objects;

[0304] Step 1304: Input the sequential samples and the non-sequential samples into the object recommendation model to be trained, and extract features from the sequential samples and the non-sequential samples in the object recommendation model to obtain initial sequential sample features and initial non-sequential sample features;

[0305] Step 1306: Enhance the features of the initial sequential sample features to obtain the sequential sample features, and enhance the features of the initial non-sequential sample features to obtain the non-sequential sample features;

[0306] Step 1308: Train the object recommendation model to be trained based on the sequential sample features and the non-sequential sample features to obtain an object recommendation model.

[0307] Among them, the sample object can be understood as the behavioral object serving as a sample, and this behavioral object can be understood as the behavioral object targeted by the user behavior of the sample user. For example, in the case where the object recommendation method is applied to the commodity recommendation scenario, the sample object can be understood as the sample commodity. In the case where the object recommendation method is applied to the scenic spot recommendation scenario, the sample object can be understood as the sample scenic spot. In the case where the object recommendation method is applied to the food recommendation scenario, the sample object can be understood as the sample food. The object recommendation model can be understood as the data processing model in the above data processing model training method.

[0308] The object recommendation model training method provided by the embodiments of this specification, after determining the sequential samples and non-sequential samples of the data processing model to be trained, extracts features from the sequential samples and non-sequential samples in the object recommendation model to be trained, and enhances the initial sequential sample features and initial non-sequential sample features obtained from the feature extraction to obtain the sequential sample features and non-sequential sample features for model training; then trains the object recommendation model to be trained based on the sequential sample features and non-sequential sample features to obtain the object recommendation model. Thus, the influence of sample data on model training is fully considered during the model training process, avoiding the problem of poor model training efficiency caused by sample data, and meeting the needs of practical applications.

[0309] The above is a schematic solution of an object recommendation model training method of this embodiment. It should be noted that the technical solution of this object recommendation model training method and the technical solution of the above data processing model training method belong to the same concept. For the details not described in detail in the technical solution of the object recommendation model training method, reference can be made to the description of the technical solution of the above data processing model training method.

[0310] Corresponding to the above method embodiments, this specification also provides an embodiment of a data processing device. The device includes:

[0311] A data receiving module, configured to receive the user behavior data of the user, where the user behavior data includes sequential behavior data and / or non-sequential behavior data;

[0312] A data processing module, configured to input the sequential behavior data and / or the non-sequential behavior data into a data processing model to obtain a data processing result corresponding to the user, wherein the data processing model is obtained by training with sequential sample features and non-sequential sample features, and the sequential sample features and the non-sequential sample features are obtained by performing feature extraction on sequential samples and non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features, and performing feature enhancement on the initial sequential sample features and the initial non-sequential sample features.

[0313] Optionally, the data processing device further includes a model training module, configured to:

[0314] Determine the sequential samples and the non-sequential samples of the data processing model to be trained, wherein the sequential samples and the non-sequential samples are user behavior data of sample users;

[0315] Input the sequential samples and the non-sequential samples into the data processing model to be trained, and perform feature extraction on the sequential samples and the non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features;

[0316] Perform feature enhancement on the initial sequential sample features to obtain the sequential sample features, and perform feature enhancement on the initial non-sequential sample features to obtain the non-sequential sample features;

[0317] Train the data processing model to be trained based on the sequential sample features and the non-sequential sample features to obtain the data processing model.

[0318] Optionally, the model training module is further configured to:

[0319] Determine similar behavior objects corresponding to the user behavior objects in the initial sequential sample features, and perform sample expansion based on the user behavior objects and the similar behavior objects to obtain expanded sequential sample features;

[0320] Perform feature enhancement on the initial sequential sample features and the expanded sequential sample features to obtain the sequential sample features of the data processing model to be trained.

[0321] The data processing model in the data processing device provided by the embodiments of this specification is obtained by training using sequential sample features and non-sequential sample features. The sequential sample features and non-sequential sample features are obtained by performing feature enhancement on the initial sequential sample features and initial non-sequential sample features obtained by extracting features from sequential samples and non-sequential samples in the data processing model. Thus, during the model training process, the influence of sample data on model training is fully considered, avoiding the problem of poor model training efficiency caused by sample data. Moreover, the trained data processing model processes the sequential behavior data and / or non-sequential behavior data included in the user behavior data and obtains a data processing result, thereby meeting the needs of actual applications through the data processing model.

[0322] The above is a schematic solution of a data processing device in this embodiment. It should be noted that the technical solution of this data processing device and the technical solution of the above data processing method belong to the same concept. For the details not described in detail in the technical solution of the data processing device, reference can be made to the description of the technical solution of the above data processing method.

[0323] Corresponding to the above method embodiment, this specification also provides an embodiment of a data processing model training device, which includes:

[0324] A sample determination module configured to determine sequential samples and non-sequential samples of the data processing model to be trained, where the sequential samples and the non-sequential samples are user behavior data of sample users;

[0325] A feature extraction module configured to input the sequential samples and the non-sequential samples into the data processing model to be trained, and perform feature extraction on the sequential samples and the non-sequential samples in the data processing model to be trained to obtain initial sequential sample features and initial non-sequential sample features;

[0326] A sample enhancement module configured to perform feature enhancement on the initial sequential sample features to obtain the sequential sample features, and perform feature enhancement on the initial non-sequential sample features to obtain the non-sequential sample features;

[0327] A model training module configured to train the data processing model to be trained based on the sequential sample features and the non-sequential sample features to obtain a data processing model.

[0328] Optionally, the sample enhancement module is further configured to:

[0329] Determine a similar behavior object corresponding to the user behavior object in the initial sequence-type sample feature, and perform sample augmentation based on the user behavior object and the similar behavior object to obtain an augmented sequence-type sample feature;

[0330] Perform feature enhancement on the initial sequence-type sample feature and the augmented sequence-type sample feature to obtain the sequence-type sample feature of the data processing model to be trained.

[0331] Optionally, the sample enhancement module is further configured to:

[0332] Determine at least two user behavior objects in the initial sequence-type sample feature;

[0333] Determine the similarity result between the target behavior object and other behavior objects among the at least two user behavior objects, where the target behavior object is any one of the at least two user behavior objects, and the other behavior objects are the other user behavior objects except the target behavior object among the at least two user behavior objects;

[0334] Determine a corresponding similar behavior object for the target behavior object based on the similarity result.

[0335] Optionally, the sample enhancement module is further configured to:

[0336] Construct a user behavior object matrix based on the at least two user behavior objects;

[0337] Obtain a similarity result matrix corresponding to the at least two user behavior objects by performing similarity calculation on the user behavior object matrix;

[0338] Determine the similarity result between the target behavior object and other behavior objects among the at least two user behavior objects through the similarity result matrix.

[0339] Optionally, the similarity result is a similarity degree;

[0340] The sample enhancement module is further configured to:

[0341] Determine a target similarity degree greater than or equal to a preset similarity threshold from the similarity degrees between the target behavior object and the other behavior objects;

[0342] Determine the other behavior objects corresponding to the target similarity degree as at least two candidate similar behavior objects of the target behavior object;

[0343] Sort the at least two candidate similar behavior objects based on the target similarity degree to obtain a candidate similar behavior object sequence;

[0344] Select a preset first number of the candidate similar behavior objects from the candidate similar behavior object sequence from top to bottom as the similar behavior objects corresponding to the target behavior object.

[0345] Optionally, the similarity result is a similarity degree.

[0346] The sample enhancement module is further configured to:

[0347] When the similarity degree between the target behavior object and the other behavior objects is less than a preset similarity threshold, determine the object type information of the target behavior object.

[0348] Determine at least two candidate similar behavior objects corresponding to the object type information from the at least two user behavior objects, and determine the object quantity of each candidate similar behavior object.

[0349] Sort the at least two candidate similar behavior objects based on the object quantity to obtain a candidate similar behavior object sequence.

[0350] Select a preset second number of the candidate similar behavior objects from the candidate similar behavior object sequence from top to bottom as the similar behavior objects corresponding to the target behavior object.

[0351] Optionally, the number of the initial sequential sample features is at least two.

[0352] The sample enhancement module is further configured to:

[0353] Determine the user behavior object information of each initial sequential sample feature in the at least two initial sequential sample features.

[0354] Based on the user behavior object information, determine a to-be-expanded sequential sample feature from the at least two initial sequential sample features.

[0355] Process the user behavior objects in the to-be-expanded sequential sample feature based on the similar behavior objects to obtain an expanded sequential sample feature.

[0356] Optionally, the sample enhancement module is further configured to:

[0357] Determine a to-be-expanded behavior object and the corresponding to-be-expanded object type information of the to-be-expanded behavior object from the at least two user behavior objects included in the to-be-expanded sequential sample feature.

[0358] Select a target similar behavior object from the similar behavior objects corresponding to the to-be-expanded behavior object, and determine the target object type information of the target similar behavior object.

[0359] Replace the to-be-expanded behavior object in the to-be-expanded sequential sample feature with the target similar behavior object, and replace the to-be-expanded object type information corresponding to the to-be-expanded behavior object in the to-be-expanded sequential sample feature with the target object type information, to obtain the expanded sequential sample feature.

[0360] Optionally, the sample enhancement module is further configured to:

[0361] Input the initial sequential sample feature into the first data processing module in the to-be-trained data processing model;

[0362] Process the initial sequential sample feature by using the feature processing unit in the first data processing module to obtain a to-be-enhanced sequential sample feature;

[0363] Determine a target data enhancement rule from at least two data enhancement rules, and perform data enhancement on the to-be-enhanced sequential sample feature by using the target data enhancement rule to obtain the sequential sample feature.

[0364] Optionally, the sample enhancement module is further configured to:

[0365] Select two target data enhancement rules from a preset first data enhancement rule, a preset second data enhancement rule, and a preset third data enhancement rule;

[0366] Perform data enhancement on the to-be-enhanced sequential sample feature by using the two target data enhancement rules to obtain a first enhanced sequential sample feature and a second enhanced sequential sample feature;

[0367] Use the first enhanced sequential sample feature and the second enhanced sequential sample feature as the sequential sample feature.

[0368] Optionally, the sample enhancement module is further configured to:

[0369] Input the initial non-sequential sample feature into the second data processing module in the to-be-trained data processing model;

[0370] Perform feature masking processing on the initial non-sequential sample feature to obtain an enhanced non-sequential sample feature;

[0371] Process the enhanced non-sequential sample feature by using the feature processing unit in the second data processing module to obtain the non-sequential sample feature.

[0372] Optionally, the model training module is further configured to:

[0373] Process the sequential sample features using the first data processing module in the data processing model to be trained, and obtain a first sample processing result;

[0374] Process the non-sequential sample features using the second data processing module in the data processing model to be trained, and obtain a second sample processing result;

[0375] Train the first data processing module based on the first sample processing result, and train the second data processing module based on the second sample processing result until the model training stop condition is reached, and obtain the data processing model.

[0376] Optionally, the second data processing module includes a non-sequential data processing module and a data prediction module;

[0377] The model training module is further configured to:

[0378] Encode the non-sequential sample features using the encoding unit in the non-sequential data processing module to obtain the non-sequential data processing result;

[0379] Process the non-sequential sample features using the data prediction module to obtain a data prediction result;

[0380] Use the non-sequential data processing result and the data prediction result as the second sample processing result.

[0381] Optionally, the model training module is further configured to:

[0382] Input the non-sequential sample features into the data prediction module, and process the non-sequential sample features using the feature processing unit in the data prediction module to obtain processed non-sequential sample features;

[0383] Input the processed non-sequential sample features into the encoding unit in the data prediction module for encoding to obtain the data prediction result.

[0384] The data processing model training device provided in the embodiments of this specification, after determining the sequential samples and non-sequential samples of the data processing model to be trained, extracts features from the sequential samples and non-sequential samples in the data processing model to be trained, and enhances the initial sequential sample features and initial non-sequential sample features obtained by feature extraction to obtain sequential sample features and non-sequential sample features for model training; then trains the data processing model to be trained based on the sequential sample features and non-sequential sample features to obtain a data processing model. Thus, the influence of sample data on model training is fully considered during the model training process, avoiding the problem of poor model training efficiency caused by sample data and meeting the needs of practical applications.

[0385] The above is a schematic solution of a data processing model training device according to this embodiment. It should be noted that the technical solution of this data processing model training device and the technical solution of the above data processing model training method belong to the same concept. For the details not described in the technical solution of the data processing model training device, reference can be made to the description of the technical solution of the above data processing model training method.

[0386] Corresponding to the above method embodiment, this specification also provides another embodiment of a data processing device. This device is applied to the cloud and includes:

[0387] A data receiving module, configured to receive the user behavior data of the user sent by the terminal, where the user behavior data includes sequential behavior data and / or non-sequential behavior data;

[0388] A data processing module, configured to input the sequential behavior data and / or the non-sequential behavior data into a data processing model to obtain the data processing result corresponding to the user, where the data processing model is trained by sequential sample features and non-sequential sample features, and the sequential sample features and the non-sequential sample features are obtained by extracting features from sequential samples and non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features, and enhancing the initial sequential sample features and the initial non-sequential sample features;

[0389] A result sending module, configured to send the data processing result to the terminal.

[0390] In one or more embodiments provided in this specification, the data processing model in another data processing device is obtained by training using sequential sample features and non-sequential sample features. The sequential sample features and non-sequential sample features are obtained by performing feature enhancement on the initial sequential sample features and initial non-sequential sample features obtained by extracting features from sequential samples and non-sequential samples in the data processing model. Thus, during the model training process, the influence of sample data on model training is fully considered, avoiding the problem of poor model training efficiency caused by sample data. Moreover, the data processing model after training processes the sequential behavior data and / or non-sequential behavior data included in the user behavior data sent by the terminal to obtain a data processing result, and sends the data processing result to the terminal, thereby meeting the needs of actual applications through the data processing model.

[0391] The above is a schematic solution of another data processing device in this embodiment. It should be noted that the technical solution of this another data processing device belongs to the same concept as the technical solution of the above another data processing method. For the details not described in detail in the technical solution of this another data processing device, reference can be made to the description of the technical solution of the above another data processing method.

[0392] Corresponding to the above method embodiment, this specification also provides an embodiment of an object recommendation model training device, which includes:

[0393] A sample determination module, configured to determine sequential samples and non-sequential samples of the object recommendation model to be trained, where the sequential samples and the non-sequential samples are user behavior data of sample users for sample objects;

[0394] A feature extraction module, configured to input the sequential samples and the non-sequential samples into the object recommendation model to be trained, and perform feature extraction on the sequential samples and the non-sequential samples in the object recommendation model to obtain initial sequential sample features and initial non-sequential sample features;

[0395] A sample enhancement module, configured to perform feature enhancement on the initial sequential sample features to obtain the sequential sample features, and perform feature enhancement on the initial non-sequential sample features to obtain the non-sequential sample features;

[0396] A model training module, configured to train the object recommendation model to be trained based on the sequential sample features and the non-sequential sample features to obtain an object recommendation model.

[0397] The object recommendation model training device provided by the embodiments of this specification, after determining the sequential samples and non-sequential samples of the data processing model to be trained, extracts features from the sequential samples and non-sequential samples in the object recommendation model to be trained, and enhances the initial sequential sample features and initial non-sequential sample features obtained by feature extraction to obtain sequential sample features and non-sequential sample features for model training; then trains the object recommendation model to be trained based on the sequential sample features and non-sequential sample features to obtain an object recommendation model. Thus, the influence of sample data on model training is fully considered during the model training process, avoiding the problem of poor model training efficiency caused by sample data and meeting the needs of practical applications.

[0398] The above is a schematic solution of an object recommendation model training device according to this embodiment. It should be noted that the technical solution of this object recommendation model training device and the technical solution of the above object recommendation model training method belong to the same concept. For the details not described in the technical solution of this object recommendation model training device, reference can be made to the description of the technical solution of the above object recommendation model training method.

[0399] Figure 14 FIG. shows a structural block diagram of a computing device 1400 according to an embodiment of this specification. The components of the computing device 1400 include, but are not limited to, a memory 1410 and a processor 1420. The processor 1420 is connected to the memory 1410 through a bus 1430, and a database 1450 is used to store data.

[0400] The computing device 1400 also includes an access device 1440 that enables the computing device 1400 to communicate via one or more networks 1460. Examples of such networks include the Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 1440 may include one or more of any type of wired or wireless network interface (e.g., a network interface card (NIC)), such as an IEEE 802.12 Wireless Local Area Network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0401] In one embodiment of the present specification, the above components of the computing device 1400 and Figure 14 other components not shown therein may also be connected to each other, for example, via a bus. It should be understood that Figure 14 the block diagram of the computing device shown is for illustrative purposes only and is not a limitation on the scope of the present specification. Those skilled in the art may add or replace other components as needed.

[0402] The computing device 1400 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smartphone), a wearable computing device (e.g., a smartwatch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 1400 may also be a mobile or stationary server. Among them, the processor 1420 is used to execute the following computer-executable instructions, and when the computer-executable instructions are executed by the processor, the steps of the above two data processing methods, data processing model training methods, or object recommendation model training methods are implemented.

[0403] The above is a schematic solution of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solutions of the above two data processing methods, data processing model training methods, and object recommendation model training methods belong to the same concept. For the details not described in detail in the technical solution of the computing device, reference can be made to the descriptions of the technical solutions of the above two data processing methods, data processing model training methods, and object recommendation model training methods.

[0404] An embodiment of this specification also provides a computer-readable storage medium that stores computer-executable instructions. When the computer-executable instructions are executed by a processor, the steps of the above two data processing methods, data processing model training methods, or object recommendation model training methods are implemented.

[0405] The above is a schematic solution of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solutions of the above two data processing methods, data processing model training methods, and object recommendation model training methods belong to the same concept. For the details not described in detail in the technical solution of the storage medium, reference can be made to the descriptions of the technical solutions of the above two data processing methods, data processing model training methods, and object recommendation model training methods.

[0406] An embodiment of this specification also provides a computer program. When the computer program is executed on a computer, the computer is made to execute the steps of the above two data processing methods, data processing model training methods, or object recommendation model training methods.

[0407] The above is a schematic solution of a computer program according to this embodiment. It should be noted that the technical solution of this computer program and the technical solutions of the above two data processing methods, data processing model training methods, and object recommendation model training methods belong to the same concept. For the details not described in detail in the technical solution of the computer program, reference can be made to the descriptions of the technical solutions of the above two data processing methods, data processing model training methods, and object recommendation model training methods.

[0408] The above describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve the desired results. Additionally, the processes depicted in the figures do not necessarily require the particular order or sequential order shown to achieve the desired result. In certain implementations, multitasking and parallel processing are also possible or may be advantageous.

[0409] The computer instructions include computer program code, which may be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, mobile hard disks, magnetic disks, optical disks, computer memories, read-only memories (ROMs), random access memories (RAMs), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0410] It should be noted that for the foregoing method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the embodiments of this specification are not limited by the described action sequence, because according to the embodiments of this specification, some steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments of this specification.

[0411] In the above embodiments, the descriptions of the respective embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0412] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The alternative embodiments do not elaborate on all the details and do not limit the invention to only the specific embodiments described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the embodiments of this specification, so that those skilled in the art can well understand and utilize this specification. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. A data processing method, comprising: Receiving user behavior data of a user, wherein the user behavior data includes sequential behavior data and / or non-sequential behavior data; Inputting the sequential behavior data and / or the non-sequential behavior data into a data processing model to obtain a data processing result corresponding to the user, wherein the data processing model is obtained by training with sequential sample features and non-sequential sample features, and the sequential sample features and the non-sequential sample features are obtained by extracting features from sequential samples and non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features, and performing feature enhancement on the initial sequential sample features and the initial non-sequential sample features.

2. The data processing method according to claim 1, before inputting the sequential behavior data and / or the non-sequential behavior data into the data processing model to obtain the data processing result corresponding to the user, further comprising: Determining the sequential samples and the non-sequential samples of the data processing model to be trained, wherein the sequential samples and the non-sequential samples are user behavior data of sample users; Inputting the sequential samples and the non-sequential samples into the data processing model to be trained, and extracting features from the sequential samples and the non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features; Performing feature enhancement on the initial sequential sample features to obtain the sequential sample features, and performing feature enhancement on the initial non-sequential sample features to obtain the non-sequential sample features; Training the data processing model to be trained based on the sequential sample features and the non-sequential sample features to obtain the data processing model.

3. The data processing method according to claim 2, the performing feature enhancement on the initial sequential sample features to obtain the sequential sample features includes: Determining similar behavior objects corresponding to the user behavior objects in the initial sequential sample features, and performing sample expansion based on the user behavior objects and the similar behavior objects to obtain expanded sequential sample features; Performing feature enhancement on the initial sequential sample features and the expanded sequential sample features to obtain the sequential sample features of the data processing model to be trained.

4. A data processing model training method, comprising: Determining sequential samples and non-sequential samples of a data processing model to be trained, wherein the sequential samples and the non-sequential samples are user behavior data of sample users; Inputting the sequential samples and the non-sequential samples into the data processing model to be trained, and extracting features from the sequential samples and the non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features; Perform feature enhancement on the initial sequence-type sample features to obtain the sequence-type sample features, and perform feature enhancement on the initial non-sequence-type sample features to obtain the non-sequence-type sample features; Based on the sequence-type sample features and the non-sequence-type sample features, train the data processing model to be trained to obtain a data processing model.

5. The data processing model training method according to claim 4, wherein the performing feature enhancement on the initial sequence-type sample features to obtain the sequence-type sample features includes: Determine similar behavior objects corresponding to the user behavior objects in the initial sequence-type sample features, and perform sample expansion based on the user behavior objects and the similar behavior objects to obtain expanded sequence-type sample features; Perform feature enhancement on the initial sequence-type sample features and the expanded sequence-type sample features to obtain the sequence-type sample features of the data processing model to be trained.

6. The data processing model training method according to claim 5, wherein the determining similar behavior objects corresponding to the user behavior objects in the initial sequence-type sample features includes: Determine at least two user behavior objects in the initial sequence-type sample features; Determine the similarity results between the target behavior object and other behavior objects among the at least two user behavior objects, wherein the target behavior object is any one of the at least two user behavior objects, and the other behavior objects are the other user behavior objects except the target behavior object among the at least two user behavior objects; Based on the similarity results, determine corresponding similar behavior objects for the target behavior object.

7. The data processing model training method according to claim 6, wherein the determining the similarity results between the target behavior object and other behavior objects among the at least two user behavior objects includes: Construct a user behavior object matrix based on the at least two user behavior objects; Obtain a similarity result matrix corresponding to the at least two user behavior objects by performing similarity calculation on the user behavior object matrix; Determine the similarity results between the target behavior object and other behavior objects among the at least two user behavior objects through the similarity result matrix.

8. The data processing model training method according to claim 6, wherein the similarity result is a similarity degree; The determining corresponding similar behavior objects for the target behavior object based on the similarity results includes: Determine target similarity degrees greater than or equal to a preset similarity threshold from the similarity degrees between the target behavior object and the other behavior objects; Determine the other behavior objects corresponding to the target similarity degrees as at least two candidate similar behavior objects of the target behavior object; Sort the at least two candidate similar behavior objects based on the target similarity degrees to obtain a candidate similar behavior object sequence; Select a preset first number of the candidate similar behavior objects from the candidate similar behavior object sequence from top to bottom as the similar behavior objects corresponding to the target behavior object.

9. The data processing model training method according to claim 6, wherein the similarity result is a similarity degree; The determining of the corresponding similar behavior object for the target behavior object based on the similarity result includes: Determining the object type information of the target behavior object when the similarity degree between the target behavior object and the other behavior objects is less than a preset similarity threshold; Determining at least two candidate similar behavior objects corresponding to the object type information from the at least two user behavior objects, and determining the object quantity of each candidate similar behavior object; Sorting the at least two candidate similar behavior objects based on the object quantity to obtain a candidate similar behavior object sequence; Selecting a preset second quantity of the candidate similar behavior objects from top to bottom in the candidate similar behavior object sequence as the similar behavior objects corresponding to the target behavior object.

10. The data processing model training method according to any one of claims 6-9, wherein the quantity of the initial sequential sample features is at least two; The obtaining of the augmented sequential sample features by augmenting samples based on the user behavior object and the similar behavior object includes: Determining the user behavior object information of each of the at least two initial sequential sample features; Determining the sequential sample features to be augmented from the at least two initial sequential sample features based on the user behavior object information; Processing the user behavior object in the sequential sample features to be augmented based on the similar behavior object to obtain the augmented sequential sample features.

11. The data processing model training method according to claim 10, wherein the processing of the user behavior object in the sequential sample features to be augmented based on the similar behavior object to obtain the augmented sequential sample features includes: Determining the behavior object to be augmented and the corresponding object type information of the behavior object to be augmented from the at least two user behavior objects included in the sequential sample features to be augmented; Selecting a target similar behavior object from the similar behavior objects corresponding to the behavior object to be augmented, and determining the target object type information of the target similar behavior object; Replacing the behavior object to be augmented in the sequential sample features to be augmented with the target similar behavior object, and replacing the object type information corresponding to the behavior object to be augmented in the sequential sample features to be augmented with the target object type information to obtain the augmented sequential sample features.

12. The data processing model training method according to claim 4, wherein the obtaining of the sequential sample features by enhancing the features of the initial sequential sample features includes: Inputting the initial sequential sample features into a first data processing module in the data processing model to be trained; Processing the initial sequential sample features by a feature processing unit in the first data processing module to obtain the sequential sample features to be enhanced; Determine a target data augmentation rule from at least two data augmentation rules, and use the target data augmentation rule to perform data augmentation on the sequence-type sample features to be augmented, so as to obtain the sequence-type sample features.

13. The data processing model training method according to claim 12, wherein the determining a target data augmentation rule from at least two data augmentation rules and using the target data augmentation rule to perform data augmentation on the sequence-type sample features to be augmented, so as to obtain the sequence-type sample features, includes: Select two target data augmentation rules from a preset first data augmentation rule, a preset second data augmentation rule, and a preset third data augmentation rule; Use the two target data augmentation rules to perform data augmentation on the sequence-type sample features to be augmented, so as to obtain a first augmented sequence-type sample feature and a second augmented sequence-type sample feature; Use the first augmented sequence-type sample feature and the second augmented sequence-type sample feature as the sequence-type sample features.

14. The data processing model training method according to claim 4, wherein the performing feature augmentation on the initial non-sequence-type sample features to obtain the non-sequence-type sample features includes: Input the initial non-sequence-type sample features into a second data processing module in the data processing model to be trained; Perform feature masking processing on the initial non-sequence-type sample features to obtain augmented non-sequence-type sample features; Use a feature processing unit in the second data processing module to process the augmented non-sequence-type sample features to obtain the non-sequence-type sample features.

15. The data processing model training method according to claim 4, wherein the training the data processing model to be trained based on the sequence-type sample features and the non-sequence-type sample features to obtain the data processing model includes: Use a first data processing module in the data processing model to be trained to process the sequence-type sample features to obtain a first sample processing result; Use a second data processing module in the data processing model to be trained to process the non-sequence-type sample features to obtain a second sample processing result; Train the first data processing module based on the first sample processing result, and train the second data processing module based on the second sample processing result until a model training stop condition is reached, so as to obtain the data processing model.

16. The data processing model training method according to claim 15, wherein the second data processing module includes a non-sequence-type data processing module and a data prediction module; The using the second data processing module in the data processing model to be trained to process the non-sequence-type sample features to obtain a second sample processing result includes: Use an encoding unit in the non-sequence-type data processing module to perform encoding processing on the non-sequence-type sample features to obtain the non-sequence-type data processing result; Use the data prediction module to process the non-sequence-type sample features to obtain a data prediction result; Use the non-sequence-type data processing result and the data prediction result as the second sample processing result.

17. The data processing model training method according to claim 16, wherein the step of using the data prediction module to process the non-sequential sample features to obtain a data prediction result includes: Inputting the non-sequential sample features into the data prediction module, and using the feature processing unit in the data prediction module to process the non-sequential sample features to obtain processed non-sequential sample features; Inputting the processed non-sequential sample features into the encoding unit in the data prediction module for encoding processing to obtain the data prediction result.

18. A data processing method applied to the cloud, including: Receiving user behavior data of a user sent by a terminal, wherein the user behavior data includes sequential behavior data and / or non-sequential behavior data; Inputting the sequential behavior data and / or the non-sequential behavior data into a data processing model to obtain a data processing result corresponding to the user, wherein the data processing model is obtained by training with sequential sample features and non-sequential sample features, and the sequential sample features and the non-sequential sample features are obtained by performing feature extraction on sequential samples and non-sequential samples in the data processing model to obtain initial sequential sample features and initial non-sequential sample features, and performing feature enhancement on the initial sequential sample features and the initial non-sequential sample features; Sending the data processing result to the terminal.

19. An object recommendation model training method, including: Determining sequential samples and non-sequential samples of a to-be-trained object recommendation model, wherein the sequential samples and the non-sequential samples are user behavior data of a sample user for a sample object; Inputting the sequential samples and the non-sequential samples into the to-be-trained object recommendation model, and performing feature extraction on the sequential samples and the non-sequential samples in the to-be-trained object recommendation model to obtain initial sequential sample features and initial non-sequential sample features; Performing feature enhancement on the initial sequential sample features to obtain the sequential sample features, and performing feature enhancement on the initial non-sequential sample features to obtain the non-sequential sample features; Training the to-be-trained object recommendation model based on the sequential sample features and the non-sequential sample features to obtain an object recommendation model.

20. A computing device, including: A memory and a processor; The memory is used for storing computer-executable instructions, and the processor is used for executing the computer-executable instructions, and when the computer-executable instructions are executed by the processor, the steps of the data processing method according to any one of claims 1 to 3, the data processing model training method according to any one of claims 4 to 17, the data processing method according to any one of claims 18, or the object recommendation model training method according to any one of claims 19 are implemented.

21. A computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, the steps of the data processing method according to any one of claims 1 to 3, the data processing model training method according to any one of claims 4 to 17, the data processing method according to any one of claims 18, or the object recommendation model training method according to any one of claims 19 are implemented.

Images