Method and device for content recommendation, equipment and storage medium

The content item embedding representation and recommendation are extracted through a hierarchical machine learning model, which solves the problem of poor content recommendation performance in the prior art, and achieves a more accurate and efficient recommendation effect.

CN119003876APending Publication Date: 2024-11-22BEIJING YOUZHUJU NETWORK TECH CO LTD +1

Patent Information

Application Number
CN202411125696.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-15
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The existing ID-based content recommendation method has poor performance in cold-start scenarios of content, and it is difficult to effectively model complex and diverse content recommendations. The application efficiency of language models in recommendation systems is low, and their potential is not fully explored.

Method used

Using a hierarchical machine learning model, the first machine learning model extracts embedded representations from the description information of the content items, and uses the second machine learning model to determine the recommended content items based on these embedded representations, reduce the input sequence length, and improve the accuracy of recommendations. .

Benefits of technology

It improves the accuracy and performance of content recommendations, achieves more effective user demand modeling, and improves the efficiency and user experience of the recommendation system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119003876A_ABST
    Figure CN119003876A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides a content recommendation method and device, equipment, a storage medium and a program product. The method comprises the following steps: acquiring a content item sequence associated with historical behavior data of a target user, wherein the content item sequence comprises a plurality of content items of which conversion behaviors are sequentially executed by the target user; using a first machine learning model to determine a plurality of content item embedded representations corresponding to the plurality of content items based on the first prompt lexical element and the description information of each content item in the plurality of content items, the first prompt lexical element indicating to extract the corresponding content item embedded representation from the description information of each content item; and determining, using a second machine learning model, a recommended content item to be recommended to the target user based at least on the plurality of content item embedded representations. In this way, the layered machine learning model is used for content recommendation, the length of the input sequence can be reduced, and the performance of the machine learning model and the recommendation accuracy are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] Example embodiments of the present disclosure generally relate to the field of computer technology, and more particularly, to methods, devices, apparatuses, and computer-readable storage media for content recommendation. Background Art

[0002] Recommendation algorithms need to understand user needs to predict future behaviors across various content items. The key to effective recommendation lies in accurately modeling content items and user features. Currently, mainstream recommendation methods are mainly based on identification (ID), which is encoded by converting content items and users into IDs and creating corresponding embedding tables. These methods use collaborative filtering to learn content items and user features. In order to capture the needs of different users for content, related technologies provide sequential modeling methods. However, these methods are usually dominated by embedding parameters, the model size is relatively small, and they rely heavily on ID features, resulting in poor performance in cold start scenarios of content. Relatively shallow neural networks have difficulty modeling complex and diverse content recommendations. Summary of the invention

[0003] In a first aspect of the present disclosure, a method for content recommendation is provided. The method includes: obtaining a content item sequence associated with historical behavior data of a target user, the content item sequence including multiple content items for which conversion behaviors are sequentially performed by the target user; using a first machine learning model, respectively based on a first prompt word and description information of each of the multiple content items, determining multiple content item embedding representations corresponding to each of the multiple content items, the first prompt word indicating that the corresponding content item embedding representation is extracted from the description information of each content item; and using a second machine learning model, at least based on the multiple content item embedding representations, determining a recommended content item to be recommended to the target user.

[0004] In a second aspect of the present disclosure, a device for content recommendation is provided. The device includes: a content item sequence acquisition module, configured to acquire a content item sequence associated with historical behavior data of a target user, the content item sequence including multiple content items for which conversion behaviors are sequentially performed by the target user; a content item embedding representation determination module, configured to use a first machine learning model to determine multiple content item embedding representations corresponding to each of the multiple content items based on a first prompt word and description information of each of the multiple content items, respectively, the first prompt word indicating that the corresponding content item embedding representation is extracted from the description information of each content item; and a recommended content item determination module, configured to use a second machine learning model to determine a recommended content item to be recommended to the target user based at least on the multiple content item embedding representations.

[0005] In a third aspect of the present disclosure, an electronic device is provided. The device includes at least one processing unit; and at least one memory, the at least one memory is coupled to the at least one processing unit and stores instructions for execution by the at least one processing unit. When the instructions are executed by the at least one processing unit, the electronic device executes the method of the first aspect.

[0006] In a fourth aspect of the present disclosure, a computer-readable storage medium is provided, wherein a computer program is stored on the medium, and when the computer program is executed by a processor, the method of the first aspect is implemented.

[0007] In a fifth aspect of the present disclosure, a computer program product is provided, wherein the computer program product comprises a computer program, and when the computer program is executed by a processor, the method of the first aspect is implemented.

[0008] It should be understood that the content described in this section is not intended to limit the key features or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, wherein:

[0010] Figure 1 A schematic diagram showing an example environment in which embodiments of the present disclosure can be implemented;

[0011] Figure 2 A schematic diagram showing the architecture of a recommendation system according to some embodiments of the present disclosure;

[0012] Figure 3A A schematic diagram showing the architecture of a recommendation system according to other embodiments of the present disclosure;

[0013] Figure 3B A schematic diagram showing the architecture of a recommendation system according to other embodiments of the present disclosure;

[0014] Figure 4 A schematic diagram showing the application of a first machine learning model and a second machine learning model to an online recommendation model according to some embodiments of the present disclosure is shown;

[0015] Figure 5 A flowchart showing a method for content recommendation according to some embodiments of the present disclosure is shown;

[0016] Figure 6 An exemplary structural block diagram of an apparatus for content recommendation according to some embodiments of the present disclosure is shown; and

[0017] Figure 7 A block diagram of an electronic device in which one or more embodiments of the present disclosure may be implemented is shown. DETAILED DESCRIPTION

[0018] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0019] In the description of the embodiments of the present disclosure, the term "including" and similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The term "some embodiments" should be understood as "at least some embodiments". Other explicit and implicit definitions may also be included below.

[0020] It is understandable that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and relevant provisions.

[0021] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0022] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested to be performed will require obtaining and using the user's personal information, so that the user can independently choose whether to provide personal information to software or hardware such as electronic devices, applications, servers or storage media that execute operations of the technical solution of the present disclosure based on the prompt message.

[0023] As an optional but non-limiting implementation, in response to receiving an active request from the user, the prompt information is sent to the user in a manner such as a pop-up window, in which the prompt information can be presented in text form. In addition, the pop-up window can also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0024] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation of the present disclosure. Other methods that meet relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0025] As used herein, the term "model" can learn the association between the corresponding input and output from the training data, so that after the training is completed, the corresponding output can be generated for a given input. The generation of the model can be based on machine learning technology. Deep learning is a machine learning algorithm that processes inputs and provides corresponding outputs by using multi-layer processing units. A neural network model is an example of a model based on deep learning. In this article, "model" may also be referred to as "machine learning model", "learning model", "machine learning network" or "learning network", and these terms are used interchangeably in this article.

[0026] A "neural network" is a machine learning network based on deep learning. A neural network is capable of processing inputs and providing corresponding outputs, and typically includes an input layer and an output layer and one or more hidden layers between the input layer and the output layer. Neural networks used in deep learning applications typically include many hidden layers, thereby increasing the depth of the network. The layers of a neural network are connected in sequence so that the output of the previous layer is provided as input to the next layer, where the input layer receives the input of the neural network and the output of the output layer serves as the final output of the neural network. Each layer of a neural network includes one or more nodes (also called processing nodes or neurons), each of which processes input from the previous layer.

[0027] Generally, machine learning can be roughly divided into three stages, namely the training stage, the testing stage, and the application stage (also called the inference stage). In the training stage, a given model can be trained using a large amount of training data, and the parameter values ​​are continuously updated iteratively until the model can obtain consistent inferences that meet the expected goals from the training data. Through training, the model can be considered to be able to learn the association between input and output (also called input-to-output mapping) from the training data. The parameter values ​​of the trained model are determined. In the testing stage, the test input is applied to the trained model to test whether the model can provide the correct output, thereby determining the performance of the model. The testing stage can sometimes be integrated into the training stage. In the application or inference stage, the trained model can be used to process the actual model input based on the parameter values ​​obtained through training to determine the corresponding model output.

[0028] Figure 1A schematic diagram of an example environment 100 in which embodiments of the present disclosure can be implemented is shown. In the environment 100, a recommendation system 110 is configured to provide one or more specific recommended content items related to one or more resources to a user (e.g., to a terminal device 130) based on a corresponding strategy. A content library 120 includes one or more content items 122-1, 122-2, ... 122-M (collectively or individually referred to as content items 122 for ease of discussion) that can be recommended.

[0029] One or more terminal devices 130-1, 130-2, 130-3, etc. (collectively or individually referred to as terminal devices 130 for ease of discussion) are associated with the recommendation system 110. The recommendation system 110 can send corresponding content items 122 to each terminal device 110 to provide to corresponding users 132-1, 132-2, 132-3, etc. (collectively or individually referred to as users 132 for ease of discussion). As an example, the recommendation system 110 can be applied to various applications, websites, web pages, and other platforms accessible to the terminal device 130.

[0030] In this article, content items refer to information or resources to be recommended, examples of which may include video content, audio content, images, animated images, text content, various physical or virtual objects, such as applications, physical goods, virtual goods, etc. In this article, an audience group may include one or more audience members, such as user 132. A user may be any potential consumer of a resource.

[0031] In environment 100, terminal device 110 can be any type of mobile terminal, fixed terminal or portable terminal, including mobile phone, desktop computer, laptop computer, notebook computer, netbook computer, tablet computer, media computer, multimedia tablet, personal communication system (PCS) device, personal navigation device, personal digital assistant (PDA), audio / video player, digital camera / camcorder, positioning device, television receiver, radio broadcast receiver, e-book device, game device or any combination of the foregoing, including accessories and peripherals of these devices or any combination thereof. In some embodiments, terminal device 110 can also support any type of interface for users (such as "wearable" circuits, etc.). Recommendation system 110 can be implemented, for example, in various types of computing systems / servers that can provide computing power, including but not limited to mainframes, edge computing nodes, computing devices in cloud environments, and the like.

[0032] It should be understood that the structure and function of the various elements in the environment 100 are described for exemplary purposes only and do not imply any limitation on the scope of the present disclosure.

[0033] As mentioned earlier, there are some shortcomings in ID-based recommendation methods. Language models (LM) have made breakthroughs in various fields, demonstrating world knowledge and reasoning capabilities. The success of language models has inspired the exploration of integrating language models into recommendation systems. These explorations can be roughly divided into three methods. The first method is to use language models to provide supplementary information to recommendation systems, such as summaries of user behaviors and expansions of content item information. The second method is to convert the recommendation system into a dialogue-driven format compatible with language models. The third method is to modify the language model to handle recommendation tasks, not just text input and output. This includes methods that input ID features into language models, and methods that replace existing recommendation neural networks with language models to directly optimize for objectives such as click-through rate.

[0034] Despite the above explorations, the integration of language models with recommendation systems still presents significant challenges in terms of complexity and effectiveness. One problem is that inputting user behavior history as text into a language model results in an input sequence that is too long. Therefore, compared to ID-based methods, language models require longer sequences to represent the same time span of user behavior, while the complexity of the self-attention module in the language model scales with the square of the sequence length. In addition, multiple forwardings are required to recommend a single content item, resulting in reduced efficiency. In terms of effectiveness, the performance improvements of existing language model-based methods over traditional methods are not significant, which raises questions about whether the potential of language models has been fully utilized.

[0035] In addition, some key issues remain underexplored. First, the actual values ​​of the weights of pre-trained language models are often considered to encapsulate world knowledge and need further investigation. While language models provide zero-shot and few-shot capabilities, their value is unclear when training on large-scale recommendation data. Second, the necessity of fine-tuning for recommendation tasks is questionable. Language models pre-trained on large-scale corpora exhibit strong world knowledge, but it remains to be seen whether further fine-tuning for recommendation tasks improves or degrades performance. Finally, the scalability of language models is a hallmark feature of proven scaling laws in other fields and needs to be verified in the context of recommender systems. While some studies have successfully verified scaling laws in the recommendation field, these models have far fewer parameters compared to language models. Whether language models with billions of parameters exhibit good scalability in the recommendation field remains an open question.

[0036] In order to solve the above problems, in an embodiment of the present disclosure, a solution for content recommendation is proposed. Specifically, a sequence of content items associated with the historical behavior data of the target user is obtained, and the sequence of content items includes multiple content items for which conversion behaviors are sequentially performed by the target user. Using a first machine learning model, multiple content item embedding representations corresponding to each of the multiple content items are determined based on a first prompt word and the description information of each of the multiple content items, respectively, and the first prompt word indicates that the corresponding content item embedding representation is extracted from the description information of each content item. Using a second machine learning model, a recommended content item to be recommended to the target user is determined based on at least multiple content item embedding representations.

[0037] According to the scheme of the present disclosure, a first machine learning model can be used to output multiple content item embedding representations representing multiple content items, and then a second machine learning model can determine the recommended content items to be recommended to the target user based on the multiple content item embedding representations. In this way, using a hierarchical machine learning model for content recommendation can reduce the length of the input sequence and improve the performance of the machine learning model and the accuracy of the recommendation.

[0038] Some example embodiments of the present disclosure will be described below with continued reference to the accompanying drawings.

[0039] Figure 2 FIG. 1 is a schematic diagram showing the architecture of the recommendation system 110 according to some embodiments of the present disclosure. Figure 2 As shown, the recommendation system 110 includes a first machine learning model 210 and a second machine learning model 220.

[0040] In some embodiments, the first machine learning model 210 and the second machine learning model 220 are language models. The recommendation system 110 can use a hierarchical language model composed of the first machine learning model 210 and the second machine learning model 220 to recommend content items 122 to users. In this way, through the powerful text understanding ability of the language model, the embedded representation representing the content item can be accurately extracted, thereby improving the accuracy of the recommendation.

[0041] In order to predict the next content item that the user is interested in, it is necessary to understand the user's behavior. First, it is necessary to obtain a content item sequence associated with the historical behavior data of the target user, which includes multiple content items in which the target user sequentially performs conversion behaviors. In one example, the conversion behavior may include clicks, likes, favorites, comments, etc., and the target is interested in content items that have performed conversion behaviors.

[0042] After obtaining the sequence of content items, a plurality of content item embedding representations corresponding to each of the plurality of content items are determined based on a first hint word 212 (represented by [ITEM]) and the description information of each of the plurality of content items, using a first machine learning model 210. The first hint word 212 indicates that a corresponding content item embedding representation 216 is extracted from the description information 214 of each content item.

[0043] I can be used to represent the description information of a content item (for example, including the title, tag, description, etc. of the content item), and U = {I1, I2, ..., I n}Description information of multiple content items in chronological order, the first machine learning model 210 and the second machine learning model 220 aim to predict the next content item I n+1 . The first machine learning model 210 has a powerful ability to understand, summarize and generalize text. The first machine learning model 210 can be used to process the description information of each content item. The input description information of the first machine learning model 210 may include a piece of text, such as the title, description, tags and other information of each content item. Typically, the first machine learning model 210 is used to predict the next word based on the context. In order to enable the first machine learning model 210 to extract the features of the content item, a first prompt word 212 may also be input into the first machine learning model 210 to instruct the first machine learning model 210 to extract the corresponding content item embedding from the description information of each content item. The output of the first machine learning model 210 may include a content item embedding representation 216, which is used to characterize the content item.

[0044] In some embodiments, for each content item in the plurality of content items, a first input sequence 213 of the first machine learning model 210 may be generated based on the first prompt word 212 and the description information 214 of the content item. A prompt word 217 may be added before the description information 214, and after the description information 214 and the prompt word 217 are input into a tokenizer, a plurality of word lemmas (m) may be obtained, and the first prompt word lemma 212 may be added after the plurality of word lemmas, thereby forming a first input sequence 213. The first input sequence 213 may be represented as {t1, t2, ..., t m ,[ITEM]}, where t1 to t m [ITEM] represents the word-gram corresponding to the prompt word 217 and the description information 214, and [ITEM] represents the first prompt word-gram 212. The hidden state from the last layer corresponding to [ITEM] can be regarded as the content item embedding representation 216.

[0045] After generating the first input sequence 213, the first input sequence 213 may be processed by the first machine learning model 210 to obtain a first output sequence 217 of the first machine learning model 210, wherein the first output sequence 217 includes a content item embedding representation 216. For the first output sequence 217, the feature at the last position of the first output sequence 217 may be determined as the content item embedding representation 216 for characterizing the content item.

[0046] In some embodiments, the first hint word 212 is placed after the description information of the content item in the first input sequence 213. In this way, placing the first hint word at the end of the input sequence enables the machine learning model to consider the entire input sequence, thereby extracting a more accurate embedding representation.

[0047] In some embodiments, a second input sequence 221 for a second machine learning model 220 may be generated based on multiple content item embedding representations. The second input sequence 221 may include multiple content item embeddings, and the second input sequence 221 may be represented as {E1, E2, ..., E n}, where E i Represents content item I i Embedding representation. Then, the second input sequence 221 can be processed by the second machine learning model 220 to obtain a second output sequence 222 of the second machine learning model. The output unit at a given position in the second output sequence 222 indicates the content item embedding representation at the given position predicted based on the content item embedding representation before the given position in the second input sequence 221. Taking the output unit E'5 224 at the fourth position in the second output sequence 222 as an example, E'4 224 indicates the content item embedding representations at four positions predicted based on the content item embedding representations before the fourth position of the second input sequence 221 (i.e., E1, E2, and E3).

[0048] After obtaining the second output sequence 222, the recommended content item for the target user may be determined based on the content item embedding representation indicated by the last output unit 223 in the second output sequence 222. The content item embedding representation indicated by the last output unit 223 may be predicted based on the embedding representations of all content items included in the second input sequence.

[0049] In some embodiments, a recommended content item may be selected from multiple candidate content items based on the similarity between the content item embedding representation indicated by the last output unit 223 and the content item embedding representation corresponding to the multiple candidate content items. For example, multiple candidate content items may be sorted in descending order according to the similarity, and the candidate content item with the highest similarity may be recommended to the target user as the recommended content item. In this way, generative recommendation may be implemented, and customized recommended content may be provided according to the needs of different users for content items, thereby improving user experience.

[0050] The embodiments of the present disclosure can not only realize generative recommendation, but also discriminative recommendation. Figure 3A To describe the discriminant recommendation provided by some embodiments of the present disclosure, Figure 3A FIG. 3 shows a schematic diagram of the architecture of the recommendation system 110 according to some other embodiments of the present disclosure. In discriminative recommendation, a candidate content item may be given first, and then it is determined whether the candidate content item is to be recommended. As shown in FIG. 3 , the architecture and processing process related to the first machine learning model 210 may refer to FIG. Figure 2 For the second machine learning model 220, a third input sequence 322 for the second machine learning model 220 can be generated based on the second prompt word 321 (represented by [USER]) and multiple content item embedding representations.

[0051] Typically, the second machine learning model 220 is used to predict the next word based on the context. In order to enable the second machine learning model 220 to extract the characteristics of the target user, a second prompt word 321 may be input to the second machine learning model 220 to instruct the second machine learning model 220 to extract a user embedding representation 323 for the target user from multiple content item embedding representations. Then, the second machine learning model 220 may be used to process the third input sequence 322 to obtain a third output sequence 324 of the second machine learning model 220, and the third output sequence 324 includes the user embedding representation 323.

[0052] After obtaining the user embedding representation 323, based on the user embedding representation 323 and the content item embedding representation of at least one candidate content item (e.g., the content item embedding representation 325), the probability of each candidate content item in the at least one candidate content item being recommended to the target user is determined. The embedding representation of the candidate content item is the content item embedding representation 325 (given by E tgtFor example, the prediction head 326 may be used to input the user embedding representation 323 and the content item embedding representation 325 into the prediction head 326, and the probability of the content item corresponding to the content item embedding representation 325 being recommended to the target user may be obtained. For example, the prediction head 326 may be a multilayer perceptron (MLP for short). Then, the recommended content item 122 may be determined from at least one candidate content item based on the probability. For example, the candidate content item with a probability greater than 50% may be determined as the recommended content item to be recommended to the target user.

[0053] In some embodiments, the second hint word 321 is placed after the multiple content item embedding representations in the third input sequence 322. In this way, placing the second hint word at the end of the third input sequence enables the machine learning model to consider the entire input sequence, thereby extracting a more accurate embedding representation.

[0054] The following will refer to Figure 3B To describe another discriminant recommendation provided by some embodiments of the present disclosure, Figure 3B FIG. 1 is a schematic diagram showing the architecture of the recommendation system 110 according to some other embodiments of the present disclosure. Figure 3B As shown, for each candidate content item in at least one candidate content item (for example, a content item corresponding to the content item embedding representation 325), a fourth input sequence 351 for the second machine learning model 220 can be generated based on the multiple content item embedding representations and the content item embedding representations of the candidate content items. Then, the fourth input sequence 351 is processed using the second machine learning model 220 to obtain a fourth output sequence 352 of the second machine learning model 220.

[0055] After obtaining the fourth output sequence 352, the recommended content item can be determined from at least one candidate content item based on the fourth output sequence 352 generated for at least one content item. The fourth output sequence 352 can include a high-order cross feature 353 that characterizes the candidate content item. The cross feature 353 is input to the prediction head 326 to obtain the probability of the candidate content item being recommended. For example, if the probability of the candidate content item is greater than 50%, the candidate content item is determined as a recommended content item.

[0056] In some embodiments, the prediction head 326 may be replaced by an online recommendation model. The online recommendation model has a more complex structure and can output a more accurate probability, thereby improving the accuracy of determining the recommended content items.

[0057] Figure 4 A schematic diagram 400 of applying the first machine learning model 210 and the second machine learning model 220 to an online recommendation model according to some embodiments of the present disclosure is shown. Figure 4As shown, if a content item is created in the content item creation flow 410, the trained first machine learning model 210 is used to generate a content item embedding representation of the content item. The generated content item embedding representation is saved to the content item storage 430. The content item storage 430 may store a series of content items that can be recommended.

[0058] For the target user to whom content is to be recommended, a content item embedding representation corresponding to a series of content items for which the user has historically performed conversion behaviors is obtained from the content item storage 430 based on the historical behaviors of the target user as the input of the second machine learning model 220. Figure 3A In the described embodiment, the trained second machine learning model 220 generates a user embedding representation based on the acquired content item embedding representation. In some embodiments, the user embedding representation of each user can be determined at a certain period, for example, based on the daily user flow 420 (e.g., including content items for which the user performs conversion activities).

[0059] The generated user embedding representation is saved to the user storage 440. After receiving the online request 450, the online recommendation model 460 can obtain the corresponding user embedding representation from the user storage 440, obtain the content item embedding of at least one candidate content item from the content item storage 430, and determine the content item to be recommended based on the user embedding representation and the content item embedding.

[0060] Understandably, Figure 4 A possible application scenario of the machine learning model based on the embodiments of the present disclosure is given. In some embodiments, reference Figure 2 The described content recommendation architecture can be directly deployed for online content recommendation. Figure 3B The described content recommendation architecture can be applied to call the second machine learning model 220 in real time to determine the probability between the target user to be recommended and each candidate machine learning model.

[0061] In some embodiments, the first machine learning model 210 and the second machine learning model 220 can be trained. Figure 2 In the recommended method discussed above, the first machine learning model 210 can be used to determine the corresponding multiple sample content item embedding representations of the multiple sample content items based on the first sample prompt word and the description information of each content item in the multiple sample content items. The first sample prompt word indicates that the corresponding sample content item embedding representation is extracted from the description information of each sample content item. Then, the second machine learning model 220 is used to process the first number of sample content item embedding representations to obtain the first sample output sequence of the second machine learning model. In one example, the first number is a smaller number, such as 150.

[0062] After obtaining multiple sample content item embedding representations and sample output sequences, a first loss function can be determined based on a sample output unit at a given position (e.g., the i-th position) of the first sample output sequence and a sample content item embedding representation at a position after the given position (e.g., the i+1-th position) of the first number of sample content item embedding representations. During the training process, contrastive learning loss (InfoCE loss) can be used as a training target. For any sample output unit E' in the sample output sequence i ,, the positive sample is E in the embedding representation of multiple sample content items i+1 , negative samples are other content items in the batch. The first loss function can be expressed as follows:

[0063]

[0064] where s represents the similarity function with a learnable temperature parameter, E j,i represents the content item embedding representation of the i-th content item of the j-th user (also referred to as the sample content item embedding representation), E' j,i represents the content item embedding representation (also called sample output unit) of the i-th content item predicted for the j-th user, N represents the data of the negative sample, and E j,i,k Indicates E' j,i The kth negative embedding of represents, b represents the total number of users in the training batch, and n represents the length of the user's historical behavior data. It should be understood that the above gives an example of the loss function, and various deformations of other loss functions can also be used in practical applications.

[0065] After the first loss function is constructed, the first machine learning model 210 and the second machine learning model 220 are trained by reducing or minimizing the value of the first loss function.

[0066] In some embodiments, after the first machine learning model 210 and the second machine learning model 220 are trained simultaneously, only the second machine learning model 220 may be trained. The trained first machine learning model 210 may be used to generate and store content item embedding representations corresponding to the content items, and in order to continue training the second machine learning model 220, the required content item embedding representations may be obtained from the repository. When the model parameters of the first machine learning model 210 remain unchanged, the second machine learning model 220 may be used to process a second number of sample content item embedding representations to obtain a second sample output sequence of the second machine learning model 220, where the second number is greater than the first number. For example, the first number is 150 and the second number is 1000.

[0067] Then, a first loss function is determined based on the sample output unit at a given position of the second sample output sequence and the sample content item embedding representation at a position after the given position of the second number of sample content item embedding representations. After determining the first loss function, the second machine learning model 220 is trained by reducing or minimizing the value of the first loss function. The first loss function may be a loss function as shown in formula (1). In some embodiments, the first machine learning model 210 and the second machine learning model 220 may be jointly trained. For discriminant recommendations (e.g., the recommendation method discussed with reference to FIG3 ), the first machine learning model 210 is used to determine the corresponding multiple sample content item embedding representations of the multiple sample content items based on the first sample prompt word and the description information of each sample content item in the multiple sample content items, respectively, and the first sample prompt word indicates that the corresponding sample content item embedding representation is extracted from the description information of each sample content item. Then, the second machine learning model 220 may be used to process the first number of sample content item embedding representations to obtain a first sample user embedding representation for the first sample user. Figure 3A Based on the model architecture shown, during training, the input of the second machine learning model 220 also includes a second sample prompt word, and the second sample prompt word indicates extracting a first sample user embedding representation for a first sample user from multiple sample content item embedding representations. Based on the first sample user embedding representation and the content item embedding representation of the first sample candidate content item, a first probability that the first sample candidate content item is recommended to the first sample user can be determined.

[0068] After determining the first probability that the first sample candidate content item is recommended to the first sample user, a second loss function may be determined based on the difference between the label of the first sample candidate content item and the first probability. A binary cross entropy loss function may be used as the second loss function, and the second loss function may be expressed as follows:

[0069]

[0070] Where y represents the label of the training sample (also known as the label of the sample candidate content item), and x represents the probability predicted by the second machine learning model 220.

[0071] After constructing the second loss function, the first machine learning model 210 and the second machine learning model 220 are trained by at least reducing or minimizing the value of the second loss function.

[0072] In some embodiments, for discriminant recommendation, the first loss function and the second loss function may be combined to obtain a third loss function to train the first machine learning model 210 and the second machine learning model 220. The prediction of the next content item may be used as an auxiliary loss. In one example, the first loss function and the second loss function may be weighted to obtain the third loss function. The third loss function may be expressed as follows:

[0073]

[0074] where λ controls the weight of the auxiliary loss.

[0075] In some embodiments, after simultaneously training the first machine learning model 210 and the second machine learning model 220, only the second machine learning model 220 may be trained. The trained first machine learning model 210 may be used to generate and store content item embedding representations corresponding to content items, and in order to continue training the second machine learning model 220, the required content item embedding representations may be obtained from the repository. While the model parameters of the first machine learning model 210 remain unchanged, the second machine learning model 220 may be used to process a second number of sample content item embedding representations to obtain a second sample user embedding representation for a second sample user, where the second number is greater than the first number. For example, the first number is 150, and the second number is 1000. Then, based on the second sample user embedding representation and the content item embedding representation of the second sample candidate content item, the probability of the second sample candidate content item being recommended to the second sample user may be determined.

[0076] Then, based on the difference between the label of the second sample candidate content item and the second probability, the second loss function can be determined. The second machine learning model 220 is trained by at least reducing or minimizing the value of the second loss function. The second loss function can be a loss function as shown in formula (2).

[0077] In some embodiments, the second machine learning model 220 may also be trained using a loss function as shown in formula (3).

[0078] In some embodiments, the embedding representations of the first prompt word 212 and the second prompt word 321 are learnable. The embedding representation of the first prompt word and / or the embedding representation of the second prompt word are determined during the training process of the first machine learning model 210 and the second machine learning model 220 for model reasoning. In this way, by updating the embedding representation of the first prompt word and / or the embedding representation of the second prompt word, the semantic information of the first prompt word and / or the second prompt word can be better captured, thereby improving the accuracy of the recommendation.

[0079] The above describes the solution for content recommendation using a hierarchical machine learning model (e.g., a language model) proposed in an embodiment of the present disclosure. In some experiments, it has been demonstrated that fine-tuning the hierarchical machine learning model for the recommendation target is effective, and fine-tuning can improve the performance of the hierarchical machine learning model. In some experiments, the scalability of the hierarchical machine learning model has been demonstrated, and as the number of parameters in the machine learning model increases, the recommendation performance can be effectively improved. In some experiments, it has been demonstrated that the performance of the hierarchical machine learning model proposed in an embodiment of the present disclosure is better than that of the traditional ID-based recommendation model. In some experiments, it has been demonstrated that the computational complexity is effectively reduced.

[0080] Figure 5 1 is a flowchart of a method 500 for content recommendation according to some embodiments of the present disclosure. The method 500 may be implemented in Figure 1 The recommendation system 110 will refer to Figure 1 The method 500 is described with reference to the environment 100 of FIG.

[0081] In block 510 , the recommendation system 110 obtains a content item sequence associated with historical behavior data of a target user, where the content item sequence includes a plurality of content items for which conversion behaviors are sequentially performed by the target user.

[0082] In box 520, the recommendation system 110 uses a first machine learning model to determine multiple content item embedding representations corresponding to the multiple content items based on the first prompt word and the description information of each content item in the multiple content items, respectively. The first prompt word indicates that the corresponding content item embedding representation is extracted from the description information of each content item.

[0083] At block 530 , the recommendation system 110 determines, using a second machine learning model, recommended content items to be recommended to the target user based at least on the plurality of content item embedding representations.

[0084] In some embodiments, determining multiple content item embedding representations corresponding to each of the multiple content items includes: for each content item in the multiple content items, generating a first input sequence for a first machine learning model based on a first prompt word and description information of the content item; processing the first input sequence using the first machine learning model to obtain a first output sequence of the first machine learning model, the first output sequence including the content item embedding representation.

[0085] In some embodiments, the first hint word-gram is placed after the description information of the content item in the first input sequence.

[0086] In some embodiments, determining recommended content items to be recommended to a target user using a second machine learning model at least based on multiple content item embedding representations includes: generating a second input sequence for the second machine learning model based on multiple content item embedding representations; processing the second input sequence using the second machine learning model to obtain a second output sequence of the second machine learning model, wherein an output unit at a given position in the second output sequence indicates an embedding representation of the content item at a given position predicted based on the embedding representation of the content item before the given position in the second input sequence; and determining a recommended content item based on the embedding representation of the content item indicated by the last output unit in the second output sequence.

[0087] In some embodiments, determining a recommended content item based on the content item embedding representation indicated by the last output unit in the second output sequence includes: selecting a recommended content item from multiple candidate content items based on the similarity between the content item embedding representation indicated by the last output unit and the content item embedding representations corresponding to the multiple candidate content items.

[0088] In some embodiments, determining recommended content items to be recommended to a target user using a second machine learning model based at least on multiple content item embedding representations includes: generating a third input sequence for the second machine learning model based on a second prompt word and multiple content item embedding representations, the second prompt word indicating extraction of a user embedding representation for the target user from multiple content item embedding representations; processing the third input sequence using the second machine learning model to obtain a third output sequence of the second machine learning model, the third output sequence including the user embedding representation; determining the probability of each candidate content item in at least one candidate content item being recommended to the target user based on the user embedding representation and the content item embedding representation of at least one candidate content item; and determining a recommended content item from at least one candidate content item based on the probability.

[0089] In some embodiments, the second hint word-gram is placed after the plurality of content item embedding representations in the third input sequence.

[0090] In some embodiments, determining recommended content items to be recommended to the target user using a second machine learning model based at least on multiple content item embedding representations includes: for each candidate content item of at least one candidate content item, generating a fourth input sequence for the second machine learning model based on multiple content item embedding representations and the content item embedding representation of the candidate content item; processing the fourth input sequence using the second machine learning model to obtain a fourth output sequence of the second machine learning model; and determining a recommended content item from at least one candidate content item based on the fourth output sequence generated for at least one content item.

[0091] In some embodiments, the first machine learning model and the second machine learning model are language models.

[0092] In some embodiments, the first machine learning model and the second machine learning model are trained by: using the first machine learning model, determining multiple sample content items from corresponding multiple sample content item embedding representations based on a first sample prompt word and description information of each sample content item in the multiple sample content items, respectively, the first sample prompt word indicating that the corresponding sample content item embedding representation is extracted from the description information of each sample content item; using the second machine learning model to process the embedding representation of the first number of sample content items to obtain a first sample output sequence of the second machine learning model; determining a first loss function based on the sample output unit at a given position of the first sample output sequence and the embedding representation of the sample content item at a position after the given position of the embedding representation of the first number of sample content items; and training the first machine learning model and the second machine learning model by reducing or minimizing the value of the first loss function.

[0093] In some embodiments, while the model parameters of the first machine learning model remain unchanged, the second machine learning model is also trained by: using the second machine learning model to process the embedding representation of a second number of sample content items to obtain a second sample output sequence of the second machine learning model, where the second number is greater than the first number; determining a first loss function based on the sample output unit of the second sample output sequence at a given position and the embedding representation of the sample content item at a position after the given position of the embedding representation of the second number of sample content items; and training the second machine learning model by reducing or minimizing the value of the first loss function.

[0094] In some embodiments, the first machine learning model and the second first machine learning model are trained by: using the first machine learning model, determining multiple sample content items from corresponding multiple sample content item embedding representations based on the first sample prompt word and the description information of each sample content item in the multiple sample content items, respectively, the first sample prompt word indicating that the corresponding sample content item embedding representation is extracted from the description information of each sample content item; using the second machine learning model to process the first number of sample content item embedding representations to obtain a first sample user embedding representation for a first sample user; determining a first probability that the first sample candidate content item is recommended to the first sample user based on the first sample user embedding representation and the content item embedding representation of the first sample candidate content item; determining a second loss function based on the difference between the label of the first sample candidate content item and the first probability; and training the first machine learning model and the second machine learning model by at least reducing or minimizing the value of the second loss function.

[0095] In some embodiments, while the model parameters of the first machine learning model remain unchanged, the second machine learning model is also trained by: processing a second number of sample content item embedding representations using the second machine learning model to obtain second sample user embedding representations for second sample users, the second number being greater than the first number; determining the probability that the second sample candidate content item is recommended to the second sample user based on the second sample user embedding representation and the content item embedding representation of the second sample candidate content item; determining a second loss function based on the difference between the label of the second sample candidate content item and the second probability; and training the second machine learning model by at least reducing or minimizing the value of the second loss function.

[0096] In some embodiments, an embedding representation of the first cue word-gram and / or an embedding representation of the second cue word-gram is determined during the training process of the first machine learning model and the second machine learning model.

[0097] The embodiments of the present disclosure also provide corresponding devices for implementing the above methods or processes. Figure 6 An exemplary structural block diagram of a content recommendation apparatus 600 according to some embodiments of the present disclosure is shown. The apparatus 600 may be implemented as or included in the recommendation system 110. Each module / component in the apparatus 600 may be implemented by hardware, software, firmware, or any combination thereof.

[0098] like Figure 6 As shown, the device 600 includes a content item sequence acquisition module 610, which is configured to acquire a content item sequence associated with the historical behavior data of the target user, and the content item sequence includes multiple content items for which the target user sequentially performs conversion behaviors. The device 600 also includes a content item embedding representation determination module 620, which is configured to use a first machine learning model to determine multiple content item embedding representations corresponding to each of the multiple content items based on a first prompt word and the description information of each of the multiple content items, respectively, and the first prompt word indicates that the corresponding content item embedding representation is extracted from the description information of each content item. The device 600 also includes a recommended content item determination module 630, which is configured to use a second machine learning model to determine the recommended content items to be recommended to the target user based on at least multiple content item embedding representations.

[0099] In some embodiments, the content item embedding representation determination module 620 is further configured to generate a first input sequence for a first machine learning model based on a first prompt word and description information of the content item for each content item among a plurality of content items; and process the first input sequence using the first machine learning model to obtain a first output sequence of the first machine learning model, wherein the first output sequence includes the content item embedding representation.

[0100] In some embodiments, the first hint word-gram is placed after the description information of the content item in the first input sequence.

[0101] In some embodiments, the recommended content item determination module 630 is further configured to generate a second input sequence for a second machine learning model based on multiple content item embedding representations; process the second input sequence using the second machine learning model to obtain a second output sequence of the second machine learning model, wherein the output unit at a given position in the second output sequence indicates the content item embedding representation at the given position predicted based on the content item embedding representation before the given position in the second input sequence; and determine the recommended content item based on the content item embedding representation indicated by the last output unit in the second output sequence.

[0102] In some embodiments, the recommended content item determination module 630 is further configured to select a recommended content item from the multiple candidate content items based on the similarity between the content item embedding representation indicated by the last output unit and the content item embedding representations corresponding to the multiple candidate content items.

[0103] In some embodiments, the recommended content item determination module 630 is further configured to generate a third input sequence for a second machine learning model based on a second prompt word and multiple content item embedding representations, the second prompt word indicating extraction of a user embedding representation for a target user from multiple content item embedding representations; process the third input sequence using the second machine learning model to obtain a third output sequence of the second machine learning model, the third output sequence including the user embedding representation; determine the probability of each candidate content item in at least one candidate content item being recommended to the target user based on the user embedding representation and the content item embedding representation of at least one candidate content item; and determine a recommended content item from at least one candidate content item based on the probability.

[0104] In some embodiments, the second hint word-gram is placed after the plurality of content item embedding representations in the third input sequence.

[0105] In some embodiments, the recommended content item determination module 630 is further configured to generate a fourth input sequence for the second machine learning model for each candidate content item in at least one candidate content item based on multiple content item embedding representations and content item embedding representations of the candidate content item; process the fourth input sequence using the second machine learning model to obtain a fourth output sequence of the second machine learning model; and determine a recommended content item from at least one candidate content item based on the fourth output sequence generated for at least one content item.

[0106] In some embodiments, the first machine learning model and the second machine learning model are language models.

[0107] In some embodiments, the first machine learning model and the second machine learning model are trained by: using the first machine learning model, determining multiple sample content items from corresponding multiple sample content item embedding representations based on a first sample prompt word and description information of each sample content item in the multiple sample content items, respectively, the first sample prompt word indicating that the corresponding sample content item embedding representation is extracted from the description information of each sample content item; using the second machine learning model to process the embedding representation of the first number of sample content items to obtain a first sample output sequence of the second machine learning model; determining a first loss function based on the sample output unit at a given position of the first sample output sequence and the embedding representation of the sample content item at a position after the given position of the embedding representation of the first number of sample content items; and training the first machine learning model and the second machine learning model by reducing or minimizing the value of the first loss function.

[0108] In some embodiments, while the model parameters of the first machine learning model remain unchanged, the second machine learning model is also trained by: using the second machine learning model to process the embedding representation of a second number of sample content items to obtain a second sample output sequence of the second machine learning model, where the second number is greater than the first number; determining a first loss function based on the sample output unit of the second sample output sequence at a given position and the embedding representation of the sample content item at a position after the given position of the embedding representation of the second number of sample content items; and training the second machine learning model by reducing or minimizing the value of the first loss function.

[0109] In some embodiments, the first machine learning model and the second first machine learning model are trained by: using the first machine learning model, determining multiple sample content items from corresponding multiple sample content item embedding representations based on the first sample prompt word and the description information of each sample content item in the multiple sample content items, respectively, the first sample prompt word indicating that the corresponding sample content item embedding representation is extracted from the description information of each sample content item; using the second machine learning model to process the first number of sample content item embedding representations to obtain a first sample user embedding representation for a first sample user; determining a first probability that the first sample candidate content item is recommended to the first sample user based on the first sample user embedding representation and the content item embedding representation of the first sample candidate content item; determining a second loss function based on the difference between the label of the first sample candidate content item and the first probability; and training the first machine learning model and the second machine learning model by at least reducing or minimizing the value of the second loss function.

[0110] In some embodiments, while the model parameters of the first machine learning model remain unchanged, the second machine learning model is also trained by: processing a second number of sample content item embedding representations using the second machine learning model to obtain second sample user embedding representations for second sample users, the second number being greater than the first number; determining the probability that the second sample candidate content item is recommended to the second sample user based on the second sample user embedding representation and the content item embedding representation of the second sample candidate content item; determining a second loss function based on the difference between the label of the second sample candidate content item and the second probability; and training the second machine learning model by at least reducing or minimizing the value of the second loss function.

[0111] In some embodiments, an embedding representation of the first cue word-gram and / or an embedding representation of the second cue word-gram is determined during the training process of the first machine learning model and the second machine learning model.

[0112] The units and / or modules included in the device 600 can be implemented in various ways, including software, hardware, firmware, or any combination thereof. In some embodiments, one or more units and / or modules can be implemented using software and / or firmware, such as machine executable instructions stored on a storage medium. In addition to or as an alternative to machine executable instructions, some or all of the units and / or modules in the device 600 can be implemented at least in part by one or more hardware logic components. As an example and not limitation, exemplary types of hardware logic components that can be used include field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0113] It should be understood that one or more steps in the above method can be performed by a suitable electronic device or a combination of electronic devices. Such an electronic device or a combination of electronic devices may include, for example, Figure 1 Recommendation system 110 in .

[0114] Figure 7 1 shows a block diagram of an electronic device 700 in which one or more embodiments of the present disclosure may be implemented. It should be understood that Figure 7 The electronic device 700 shown is merely exemplary and should not constitute any limitation on the functionality and scope of the embodiments described herein. Figure 7 The electronic device 700 shown can be used to implement Figure 1 Recommendation system 110 or Figure 6 Device 600.

[0115] like Figure 7As shown, the electronic device 700 is in the form of a general electronic device. The components of the electronic device 700 may include, but are not limited to, one or more processors or processing units 710, a memory 720, a storage device 730, one or more communication units 740, one or more input devices 750, and one or more output devices 760. The processing unit 710 may be an actual or virtual processor and is capable of performing various processes according to a program stored in the memory 720. In a multi-processor system, multiple processing units execute computer executable instructions in parallel to improve the parallel processing capability of the electronic device 700.

[0116] The electronic device 700 typically includes a plurality of computer storage media. Such media may be any accessible media that is accessible to the electronic device 700, including but not limited to volatile and non-volatile media, removable and non-removable media. The memory 720 may be a volatile memory (e.g., registers, caches, random access memory (RAM)), a non-volatile memory (e.g., a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), flash memory), or some combination thereof. The storage device 730 may be a removable or non-removable medium, and may include a machine-readable medium, such as a flash drive, a disk, or any other medium, which may be capable of being used to store information and / or data and may be accessed within the electronic device 700.

[0117] The electronic device 700 may further include additional removable / non-removable, volatile / non-volatile storage media. Figure 7 As shown in , a disk drive for reading or writing from a removable, non-volatile disk (e.g., a "floppy disk") and an optical drive for reading or writing from a removable, non-volatile optical disk may be provided. In these cases, each drive may be connected to the bus (not shown) by one or more data media interfaces. The memory 720 may include a computer program product 725 having one or more program modules that are configured to perform various methods or actions of various embodiments of the present disclosure.

[0118] The communication unit 740 implements communication with other electronic devices through a communication medium. Additionally, the functions of the components of the electronic device 700 can be implemented with a single computing cluster or multiple computing machines that can communicate through a communication connection. Therefore, the electronic device 700 can operate in a networked environment using a logical connection with one or more other servers, a network personal computer (PC), or another network node.

[0119] The input device 750 may be one or more input devices, such as a mouse, a keyboard, a tracking ball, etc. The output device 760 may be one or more output devices, such as a display, a speaker, a printer, etc. The electronic device 700 may also communicate with one or more external devices (not shown) through the communication unit 740 as needed, such as a storage device, a display device, etc., communicate with one or more devices that allow a user to interact with the electronic device 700, or communicate with any device that allows the electronic device 700 to communicate with one or more other electronic devices (e.g., a network card, a modem, etc.). Such communication may be performed via an input / output (I / O) interface (not shown).

[0120] According to an exemplary implementation of the present disclosure, a computer-readable storage medium is provided, on which computer-executable instructions are stored, wherein the computer-executable instructions are executed by a processor to implement the method described above. According to an exemplary implementation of the present disclosure, a computer program product is also provided, which is tangibly stored on a non-transitory computer-readable medium and includes computer-executable instructions, and the computer-executable instructions are executed by a processor to implement the method described above.

[0121] Various aspects of the present disclosure are described herein with reference to the flowcharts and / or block diagrams of the methods, devices, equipment, and computer program products implemented according to the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0122] These computer-readable program instructions can be provided to a processing unit of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processing unit of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0123] Computer-readable program instructions can be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, so that the instructions executed on the computer, other programmable data processing apparatus, or other device implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0124] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple implementations of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of a module, program segment or instruction includes one or more executable instructions for realizing the logical function of the specification. In some implementations as an update, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the function involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be realized by a special hardware-based system that performs the function or action of the specification, or can be realized by a combination of special hardware and computer instructions.

[0125] The above descriptions of various implementations of the present disclosure are exemplary, non-exhaustive, and not limited to the disclosed implementations. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described implementations. The selection of terms used herein is intended to best explain the principles of the implementations, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the various implementations disclosed herein.

Claims

1. A content recommendation method, comprising: Acquire a content item sequence associated with historical behavior data of a target user, wherein the content item sequence includes a plurality of content items for which conversion behaviors are sequentially performed by the target user; Determine, using a first machine learning model, a plurality of content item embedding representations corresponding to each of the plurality of content items based on a first prompt word and description information of each of the plurality of content items, wherein the first prompt word indicates extracting a corresponding content item embedding representation from the description information of each content item; as well as Determine, using a second machine learning model, recommended content items to be recommended to the target user based at least on the multiple content item embedding representations.

2. The method according to claim 1, wherein determining a plurality of content item embedded representations corresponding to each of the plurality of content items comprises: For each content item in the plurality of content items, generating a first input sequence for the first machine learning model based on the first prompt word and the description information of the content item; The first input sequence is processed using the first machine learning model to obtain a first output sequence of the first machine learning model, wherein the first output sequence includes a content item embedding representation. 3 . The method according to claim 2 , wherein the first hint word-gram is placed after the description information of the content item in the first input sequence.

4. The method according to claim 1, wherein determining, using a second machine learning model, based at least on the plurality of content item embedding representations, the recommended content items to be recommended to the target user comprises: generating a second input sequence for the second machine learning model based on the plurality of content item embedding representations; Processing the second input sequence using the second machine learning model to obtain a second output sequence of the second machine learning model, wherein an output unit at a given position in the second output sequence indicates an embedding representation of a content item at the given position predicted based on an embedding representation of a content item before the given position in the second input sequence; as well as The recommended content item is determined based on the embedded representation of the content item indicated by the last output unit in the second output sequence.

5. The method according to claim 4, wherein determining the recommended content item based on the content item embedding representation indicated by the last output unit in the second output sequence comprises: The recommended content item is selected from the multiple candidate content items based on the similarity between the content item embedding representation indicated by the last output unit and the content item embedding representations corresponding to the multiple candidate content items.

6. The method according to claim 1, wherein determining, using a second machine learning model, based at least on the plurality of content item embedding representations, the recommended content items to be recommended to the target user comprises: generating a third input sequence for the second machine learning model based on a second prompt word-gram and the plurality of content item embedding representations, the second prompt word-gram indicating extracting a user embedding representation for the target user from the plurality of content item embedding representations; Processing the third input sequence using the second machine learning model to obtain a third output sequence of the second machine learning model, wherein the third output sequence includes the user embedding representation; determining, based on the user embedding representation and the content item embedding representation of at least one candidate content item, a probability of each candidate content item in the at least one candidate content item being recommended to the target user; as well as Based on the probability, the recommended content item is determined from the at least one candidate content item. 7 . The method according to claim 6 , wherein the second hint word-gram is placed after the plurality of content item embedding representations in the third input sequence.

8. The method according to claim 1, wherein determining, using a second machine learning model, based at least on the plurality of content item embedding representations, the recommended content items to be recommended to the target user comprises: for each candidate content item of at least one candidate content item, generating a fourth input sequence for the second machine learning model based on the plurality of content item embedding representations and the content item embedding representation of the candidate content item; Processing the fourth input sequence using the second machine learning model to obtain a fourth output sequence of the second machine learning model; as well as The recommended content item is determined from the at least one candidate content item based on a fourth output sequence generated for the at least one candidate content item.

9. The method of claim 1, wherein the first machine learning model and the second machine learning model are language models.

10. The method of claim 1, wherein the first machine learning model and the second machine learning model are trained by: Determine, using the first machine learning model, based on a first sample hint word and description information of each of the plurality of sample content items, embedding representations of the plurality of sample content items corresponding to the plurality of sample content items, wherein the first sample hint word indicates extracting a corresponding sample content item embedding representation from the description information of each sample content item; Processing a first number of sample content item embedding representations using the second machine learning model to obtain a first sample output sequence of the second machine learning model; determining a first loss function based on a sample output unit at a given position of the first sample output sequence and a sample content item embedding representation at a position after the first number of sample content item embedding representations of the given position; as well as The first machine learning model and the second machine learning model are trained by reducing or minimizing the value of the first loss function.

11. The method according to claim 10, wherein when the model parameters of the first machine learning model remain unchanged, the second machine learning model is further trained by: Processing a second number of sample content item embedding representations using the second machine learning model to obtain a second sample output sequence of the second machine learning model, the second number being greater than the first number; determining the first loss function based on a sample output unit at a given position of the second sample output sequence and a sample content item embedding representation of the second number of sample content item embedding representations at a position after the given position; as well as The second machine learning model is trained by reducing or minimizing the value of the first loss function.

12. The method of claim 1, wherein the first machine learning model and the second first machine learning model are trained by: Determine, using the first machine learning model, based on a first sample hint word and description information of each sample content item in the plurality of sample content items, embedding representations of the plurality of sample content items corresponding to the plurality of sample content items, wherein the first sample hint word indicates extracting a corresponding sample content item embedding representation from the description information of each sample content item; Processing a first number of sample content item embedding representations using the second machine learning model to obtain a first sample user embedding representation for a first sample user; determining, according to the first sample user embedding representation and the content item embedding representation of the first sample candidate content item, a first probability that the first sample candidate content item is recommended to the first sample user; determining a second loss function based on a difference between the label of the first sample candidate content item and the first probability; as well as The first machine learning model and the second machine learning model are trained by at least reducing or minimizing the value of the second loss function.

13. The method according to claim 12, wherein when the model parameters of the first machine learning model remain unchanged, the second machine learning model is further trained by: Processing a second number of sample content item embedding representations using the second machine learning model to obtain second sample user embedding representations for second sample users, the second number being greater than the first number; determining, according to the second sample user embedding representation and the content item embedding representation of the second sample candidate content item, a probability that the second sample candidate content item is recommended to the second sample user; determining the second loss function based on a difference between the label of the second sample candidate content item and the second probability; as well as The second machine learning model is trained by at least reducing or minimizing a value of the second loss function.

14. A method according to claim 10 or 12, wherein the embedding representation of the first prompt word and / or the embedding representation of the second prompt word is determined during the training process of the first machine learning model and the second machine learning model.

15. A device for content recommendation, comprising: A content item sequence acquisition module is configured to acquire a content item sequence associated with historical behavior data of a target user, wherein the content item sequence includes a plurality of content items for which conversion behaviors are sequentially performed by the target user; a content item embedding representation determining module, configured to determine, using a first machine learning model, a plurality of content item embedding representations corresponding to each of the plurality of content items based on a first hint word and description information of each of the plurality of content items, wherein the first hint word indicates that a corresponding content item embedding representation is extracted from the description information of each content item; as well as The recommended content item determination module is configured to use a second machine learning model to determine the recommended content items to be recommended to the target user based at least on the multiple content item embedding representations.

16. An electronic device, comprising: at least one processing unit; as well as At least one memory, the at least one memory being coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions causing the apparatus to perform a method according to any one of claims 1 to 14 when executed by the at least one processing unit.

17. A computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the method according to any one of claims 1 to 14 when executed by a processor.

18. A computer program product, comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 14.

Citation Information

Patent Citations

  • Object recommendation method and device, electronic equipment and readable storage medium

    CN116932884A

  • Content recommendation method and device for generating unstructured tags based on large language model

    CN118152658A

  • Recommendation model, method, device and equipment

    CN118227881A

  • Content recommendation model training method, content recommendation method and related equipment

    CN118245669A

  • Multi-modal sequence-oriented recommendation method

    CN118261670A

Cited By

  • Content processing method and device, equipment and storage medium

    CN120296770A

  • Personalized multimodal analysis for content item recommendation

    US12563267B1