A semantic-based meta-comparative sequence recommendation method

Abstract

The present application relates to the technical field of machine learning, and in particular to a semantic-based meta-comparative sequence recommendation method. The method comprises: using an encoder to learn the sequence representation of a user's item interaction sequence; performing a first comparative learning on a given item interaction sequence and the sequence representation of a positive sample learned by the encoder; determining the probability of the next item that the user may interact with based on the sequence representation of the item interaction sequence; using a semantic regularizer to generate a corresponding semantic representation based on the sequence representation of the item interaction sequence; generating a corresponding corrected representation based on the semantic representation and the sequence representation of a given item interaction sequence and a positive sample; performing a second comparative learning based on the sequence representation and the corrected representation of a given item interaction sequence and a positive sample; and using a meta-update strategy to train and update the encoder and regularizer to generate a sequence recommendation model. The present application provides a semantic-based meta-comparative sequence recommendation method with high item recommendation accuracy.

Description

Technical Field

[0001] The present application relates to the technical field of machine learning, and in particular to a semantic-based meta-comparison sequence recommendation method. Background Art

[0002] A recommendation system is a system that recommends items of interest to users based on their historical interaction information. Contrastive learning can alleviate the data sparsity and noise problems in the field of sequence recommendation, but its performance depends largely on information-rich features. However, the existing sequence recommendation methods based on contrastive learning mainly rely on random enhancement to create contrast pairs. This use of random enhancement only at the model level will discard some important neurons, which may cause the semantics of the enhanced view to be different from the original sequence, resulting in poor recommendation results. Summary of the invention

[0003] In order to solve the deficiencies in the prior art, the purpose of the present application is to provide a semantic-based meta-comparison sequence recommendation method to improve the accuracy of sequence recommendation.

[0004] To achieve the above objectives, the present application provides a semantic-based meta-comparison sequence recommendation method, comprising:

[0005] Use an encoder to learn a sequence representation of a user's item interaction sequence;

[0006] Any item interaction sequence in the item interaction sequence that has the same target item as the given item interaction sequence is taken as a positive sample, and the given item interaction sequence and the sequence representation learned by the encoder are subjected to a first comparative learning;

[0007] Based on the sequence representation of the item interaction sequence, determine the probability of the next item that the user may interact with;

[0008] Based on the sequence representation of the item interaction sequence, a semantic regularizer is used to generate the corresponding semantic representation;

[0009] Generate a corresponding revised representation based on the semantic representation and a given item interaction sequence and a sequence representation of a positive sample;

[0010] Based on the sequence representation and the corrected representation of the given item interaction sequence and positive samples, a second contrastive learning is performed;

[0011] The encoder and regularizer are updated using a meta-update strategy to generate a sequential recommendation model.

[0012] Furthermore, the specific step of using the encoder to learn the sequence representation of the user's item interaction sequence adopts the following formula:

[0013] ;

[0014] in, is a sequence representation, For the encoder, are model parameters, represents the embedding of items in the item interaction sequence in the vector space, represents the embedding of item interaction sequence in vector space, Indicates the position embedding of items in vector space, sequence representation is the last vector in the embedding of the item interaction sequence in the vector space , is the dimension of the vector space, is the length of the item interaction sequence, Represents a user.

[0015] Furthermore, the specific steps of taking any item interaction sequence in the item interaction sequence that has the same target item as the given item interaction sequence as a positive sample and performing comparative learning on the given item interaction sequence and the sequence representation learned by the encoder of the positive sample include:

[0016] Among the item interaction sequences in the same training batch, any item interaction sequence with the same target item as the given item interaction sequence and the given item interaction sequence are taken as positive sample pairs, and the remaining item interaction sequences in the same training batch are taken as negative samples, and the sequence representations learned by the encoder of the positive sample pairs are compared and learned.

[0017] Furthermore, the specific step of determining the probability of the next item that the user may interact with based on the sequence representation of the item interaction sequence adopts the following formula:

[0018] ;

[0019] ;

[0020] in, is a sequence representation, For item set The embedding matrix of , i is the item, represents the prediction scores of all items, represents the true value of the item interaction sequence, The recommended loss.

[0021] Furthermore, the calculation formula of the semantic regularizer is as follows:

[0022] ;

[0023] in, For semantic representation, is the semantic regularizer, are the parameters of the semantic regularizer.

[0024] Furthermore, the semantic regularizer is composed of a multi-layer perceptron and a GELU activation function.

[0025] Furthermore, the specific step of generating the corresponding modified representation based on the semantic representation and the given item interaction sequence and the sequence representation of the positive sample adopts the following formula:

[0026] ;

[0027] ;

[0028] in, and Represent the sequence representation of the given item interaction sequence and positive sample respectively, For semantic representation, and They represent the revised representations of the given item interaction sequence and positive samples respectively, and ⨀ represents the dot product.

[0029] Furthermore, the specific steps of using the meta-update strategy to train and update the encoder and the regularizer to generate the sequence recommendation model include:

[0030] Initialize the encoder and semantic regularizer, and back-propagate to update the encoder according to the recommendation loss, the contrastive loss of the first contrastive learning, and the recommendation loss of the second contrastive learning;

[0031] The encoder parameters are fixed, the contrastive loss for the second contrastive learning is calculated based on the newly encoded sequence representation, and the semantic regularizer is updated by back-propagation.

[0032] To achieve the above purpose, the electronic device provided by the present application includes:

[0033] processor;

[0034] a memory having stored thereon one or more computer program instructions executed on the processor;

[0035] When the processor runs the computer instructions, the above-mentioned semantic-based meta-comparison sequence recommendation method is executed.

[0036] To achieve the above objectives, the present application provides a computer-readable storage medium on which computer instructions are stored. When the computer instructions are executed by a processor, the steps of the semantic-based meta-comparison sequence recommendation method described above are executed.

[0037] The present application discloses a semantic-based meta-contrastive sequence recommendation method, which extracts original semantics from sequence representation and uses the captured semantic information to adaptively correct the randomly enhanced sequence representation. At the same time, a meta-update method is introduced to alleviate the negative impact of the gap between multiple tasks and further improve the performance of contrastive learning, with high recommendation accuracy.

[0038] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent from the description, or may be understood by practicing the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings are used to provide a further understanding of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings:

[0040] Figure 1 A flowchart of the semantic-based meta-comparison sequence recommendation method of the present application;

[0041] Figure 2 This is a schematic diagram of the structure of the sequence recommendation model of this application;

[0042] Figure 3 This is a schematic diagram of the structure of the semantic regularizer of this application;

[0043] Figure 4 This is a schematic diagram of the algorithm of the meta-update strategy of this application;

[0044] Figure 5 A schematic diagram of the performance indicators of the item recommendation model of the present application. DETAILED DESCRIPTION

[0045] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as being limited to the embodiments described herein. Instead, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are only for exemplary purposes and are not intended to limit the scope of protection of the present application.

[0046] It should be understood that the various steps described in the method implementation of the present application can be performed in different orders and / or performed in parallel. In addition, the method implementation may include additional steps and / or omit the steps shown. The scope of the present application is not limited in this respect.

[0047] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0048] It should be noted that the modifications of "one" and "plurality" mentioned in this application are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more". "Plurality" should be understood as two or more.

[0049] Hereinafter, embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0050] Example 1

[0051] An embodiment of the present application provides a semantic-based meta-comparison sequence recommendation method, which will be referred to below. Figure 1-Figure 5 The semantic-based meta-comparison sequence recommendation method of the present application is described in detail.

[0052] Step S101: Use an encoder to learn a sequence representation of a user's item interaction sequence.

[0053] It should be noted that sequential recommendation aims to recommend items that the next user may interact with based on the user's historical interaction data.

[0054] For example, the user set and item set are and ,user There is an item interaction sequence that has been interacted with before ,in ) indicates user In the sequence Items that interact with the location, Represents the length of the item interaction sequence. Given the history of item interaction sequences , the goal of sequence recommendation is to Recommended users may be in The interactive items are defined as:

[0055] .

[0056] In this embodiment, the item collection The embedding matrix of the generated item is embedded into the same vector space .

[0057] In this embodiment, the specific steps of using the encoder to learn the sequence representation of the user's item interaction sequence are as follows:

[0058] ;

[0059] in, is a sequence representation, For the encoder, are model parameters, represents the embedding of items in the item interaction sequence in the vector space, represents the embedding of item interaction sequence in vector space, Indicates the position embedding of items in vector space, sequence representation Embedding of item interaction sequence in vector space The last vector in , is the dimension of the vector space, is the length of the item interaction sequence, Represents a user.

[0060] Step S102: taking any item interaction sequence in the item interaction sequence that has the same target item as the given item interaction sequence as a positive sample, and performing a first comparative learning on the given item interaction sequence and the sequence representation learned by the encoder of the positive sample.

[0061] In this implementation, a supervised sampling strategy is adopted. For a given item interaction sequence , all with A sequence of item interactions with the same target item is randomly selected as its positive sample and recorded as , generated by the encoder and The corresponding sequence representation and ,right and Perform contrastive learning, and the contrastive learning loss is as follows:

[0062] ;

[0063] .

[0064] in, represents the embedding of a pair of positive samples, s(∙) represents the inner product, and neg represents the set of negative sample embeddings. For a given batch size , the same training batch In the item interaction sequences of , any item interaction sequence with the given item interaction sequence having the same target item and the given item interaction sequence are taken as positive sample pairs, and the remaining 2(|B|-1) item interaction sequences in the same training batch are taken as negative samples, and the sequence representations learned by the encoder of the positive sample pairs are compared and learned.

[0065] Step S103: determining the probability of the next item that the user may interact with based on the sequence representation of the item interaction sequence;

[0066] The specific formula is as follows:

[0067] ;

[0068] ;

[0069] in, is a sequence representation, For item set The embedding matrix of , i is the item, represents the prediction scores of all items, represents the true value of the item interaction sequence, The recommended loss.

[0070] Step S104: Based on the sequence representation of the item interaction sequence, a semantic regularizer is used to generate a corresponding semantic representation.

[0071] In this implementation, in order to retain more semantic information in the sequence representation for contrastive learning, a simple multi-layer perceptron and GELU activation function are used to construct a semantic regularizer to capture the sequence representation. The original semantic information in is calculated as follows:

[0072] ;

[0073] in, For semantic representation, is the semantic regularizer, are the parameters of the semantic regularizer.

[0074] See also Figure 3 , Figure 3 Four semantic regularizers with different structures are shown, where FC represents a simple linear layer.

[0075] Step S105: Generate a corresponding revised representation based on the semantic representation and the given object interaction sequence and the sequence representation of the positive sample.

[0076] The specific steps use the following formula:

[0077] ;

[0078] ;

[0079] in, and Represent the sequence representation of the given item interaction sequence and positive sample respectively, For semantic representation, and They represent the revised representations of the given item interaction sequence and positive samples respectively, and ⨀ represents the dot product.

[0080] Step S106: Perform a second comparative learning based on the given item interaction sequence and the sequence representation and the revised representation of the positive sample.

[0081] The calculation formula of the contrast loss of the second contrastive learning is as follows:

[0082] .

[0083] Step S107: Use the meta-update strategy to train and update the encoder and regularizer to generate a sequence recommendation model.

[0084] Since the training needs to update the parameters of the encoder and semantic regularizer, and the encoder and semantic regularizer have different goals, the encoder aims to learn better representations for the recommendation task, while the regularizer aims to capture more appropriate semantic features for the auxiliary comparison task. Therefore, in this implementation, a meta-update strategy is used for training update, see Figure 4 ,include:

[0085] Phase 1: Initializing the encoder and semantic regularizer Parameters; According to the recommendation loss, the contrast loss of the first contrastive learning and the recommendation loss of the second contrastive learning, the total loss is determined as follows: ;in is the total loss, and is a hyperparameter that needs to be tuned; by minimizing the total loss , back propagation updates the encoder to obtain the parameters learned by back propagation in the first stage .

[0086] Phase 2: Fixing the encoder parameters , according to the newly encoded sequence representation, calculate the contrast loss of the second contrastive learning , the contrastive loss of the second contrastive learning As the new contrast loss , by minimizing the contrast loss Back propagation updates the semantic regularizer and obtains the parameters of the semantic regularizer .

[0087] Repeat the first and second phases until the encoder and semantic regularizer The parameters of converge and we get the sequence recommendation model.

[0088] In this implementation, although the semantic regularizer is used, the complexity of the entire method is still The time complexity of the whole method can be considered as two stages. In the first stage, we update the parameters of the encoder using three loss functions, and the time complexity is In the second stage, we use a loss function to update the semantic regularizer with a time complexity of During the training phase, the total time complexity is ; During the testing phase, we only use the encoder, and its time complexity is exactly the same as the SASRec model.

[0089] See also Figure 5 , Figure 5 A schematic diagram of the performance indicators of the item recommendation model of this application, such as Figure 5 As shown in Figure 2, the evaluation indicators (HR@k and NDCG@k, k∈{5, 10, 20}) compare the performance of the sequence recommendation model MRCSRec of this application with all baseline models. Figure 5 The experimental results of all models on the three datasets of Sports, Beauty and Yelp are shown, from which we can see:

[0090] The sequence recommendation model MRCSRec of this application significantly outperforms the baseline model in all indicators of different datasets. For example, the HR and NDCG of MRCSRec on three datasets are improved by 1.39-4.12% and 1.28-4.47% respectively compared with the best baseline model.

[0091] Example 2

[0092] In this embodiment, an electronic device is also provided, including a processor and a memory. The memory is used to store non-temporary computer-readable instructions. The processor is used to run the non-temporary computer-readable instructions, and when the non-temporary computer-readable instructions are run by the processor, one or more steps of the semantic-based meta-comparison sequence recommendation method described above can be executed. The memory and the processor can be interconnected through a bus system and / or other forms of connection mechanisms.

[0093] For example, the processor can be a central processing unit (CPU), a digital signal processor (DSP), or other forms of processing units with data processing capabilities and / or program execution capabilities, such as a field programmable gate array (FPGA); for example, the central processing unit (CPU) can be an X86 or ARM architecture, etc.

[0094] For example, the memory may include any combination of one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, erasable programmable read-only memory (EPROM), portable compact disk read-only memory (CD-ROM), USB memory, flash memory, etc. One or more computer program modules may be stored on the computer-readable storage medium, and the processor may run one or more computer program modules to implement various functions of the electronic device. Various applications and various data, as well as various data used and / or generated by the application, etc. may also be stored in the computer-readable storage medium.

[0095] It should be noted that, in the embodiments of the present application, the specific functions and technical effects of the electronic device can refer to the above description of the semantic-based meta-comparison sequence recommendation method, which will not be repeated here.

[0096] Example 3

[0097] In this embodiment, a computer-readable storage medium is further provided, and the storage medium is used to store non-transitory computer-readable instructions. For example, when the non-transitory computer-readable instructions are executed by a computer, one or more steps in the semantic-based meta-comparison sequence recommendation method described above can be executed.

[0098] For example, the storage medium can be applied to the above-mentioned electronic device. For example, the storage medium can be the memory in the electronic device of Example 2. For example, the relevant description of the storage medium can refer to the corresponding description of the memory in the electronic device of Example 2, which will not be repeated here.

[0099] It should be noted that the storage medium (computer-readable medium) mentioned above in the present application may be a computer-readable signal medium or a non-transitory computer-readable storage medium or any combination of the above two. The non-transitory computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of non-transitory computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0100] In the present application, a non-transitory computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, apparatus, or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a non-transitory computer-readable storage medium, which may send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0101] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0102] Computer program code for performing the operations of the present application may be written in one or more programming languages, or a combination thereof, including, but not limited to, object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server.

[0103] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions 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 implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0104] The units involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of a unit does not, in some cases, constitute a limitation on the unit itself.

[0105] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may 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), etc.

[0106] The above description is only a partial embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

[0107] In addition, although each operation is described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the application. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0108] Although the subject matter has been described in language specific to structural features and / or methodological logical actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims.

Claims

1. A semantic-based meta-comparison sequence recommendation method, comprising: Use an encoder to learn a sequence representation of a user's item interaction sequence; Any item interaction sequence in the item interaction sequence that has the same target item as the given item interaction sequence is taken as a positive sample, and the given item interaction sequence and the sequence representation learned by the encoder are subjected to a first comparative learning; Based on the sequence representation of the item interaction sequence, determine the probability of the next item that the user may interact with; Based on the sequence representation of the item interaction sequence, a semantic regularizer is used to generate the corresponding semantic representation; Generate a corresponding revised representation based on the semantic representation and a given item interaction sequence and a sequence representation of a positive sample; Based on the sequence representation and the corrected representation of the given item interaction sequence and positive samples, a second contrastive learning is performed; Use the meta-update strategy to train and update the encoder and regularizer to generate a sequence recommendation model. The specific steps include: In the first stage, the encoder and semantic regularizer are initialized, and the encoder is updated by back-propagation according to the recommendation loss, the contrastive loss of the first contrastive learning, and the recommendation loss of the second contrastive learning; In the second stage, the encoder parameters are fixed, the contrastive loss of the second contrastive learning is calculated according to the newly encoded sequence representation, and the semantic regularizer is updated through back-propagation; In the third stage, the first and second stages are repeated until the parameters of the encoder and semantic regularizer converge to obtain the sequence recommendation model.

2. The semantic-based meta-comparison sequence recommendation method according to claim 1, characterized in that: The specific steps of using the encoder to learn the sequence representation of the user's item interaction sequence are as follows: ； in, is a sequence representation, For the encoder, are model parameters, represents the embedding of items in the item interaction sequence in the vector space, represents the embedding of item interaction sequence in vector space, Indicates the position embedding of items in vector space, sequence representation is the last vector in the embedding of the item interaction sequence in the vector space , is the dimension of the vector space, is the length of the item interaction sequence, Represents a user.

3. The semantic-based meta-comparison sequence recommendation method according to claim 1, characterized in that: The specific steps of taking any item interaction sequence in the item interaction sequence that has the same target item as the given item interaction sequence as a positive sample and performing comparative learning on the given item interaction sequence and the sequence representation learned by the encoder of the positive sample include: Among the item interaction sequences in the same training batch, any item interaction sequence with the same target item as the given item interaction sequence and the given item interaction sequence are taken as positive sample pairs, and the remaining item interaction sequences in the same training batch are taken as negative samples, and the sequence representations learned by the encoder of the positive sample pairs are compared and learned.

4. The semantic-based meta-comparison sequence recommendation method according to claim 1, characterized in that: The specific step of determining the probability of the next item that the user may interact with based on the sequence representation of the item interaction sequence adopts the following formula: ； ； in, is a sequence representation, For item set The embedding matrix of , i is the item, represents the prediction scores of all items, represents the true value of the item interaction sequence, The recommended loss.

5. The semantic-based meta-comparison sequence recommendation method according to claim 1, characterized in that: The calculation formula of the semantic regularizer is as follows: ； in, For semantic representation, is the semantic regularizer, are the parameters of the semantic regularizer, Represents a sequence.

6. The semantic-based meta-comparison sequence recommendation method according to claim 1, characterized in that: The semantic regularizer is composed of a multi-layer perceptron and a GELU activation function.

7. The semantic-based meta-comparison sequence recommendation method according to claim 1, characterized in that: The specific step of generating the corresponding modified representation based on the semantic representation and the given item interaction sequence and the sequence representation of the positive sample adopts the following formula: ； ； in, and Represent the sequence representation of the given item interaction sequence and positive sample respectively, For semantic representation, and Represent the corrected representation of the given item interaction sequence and positive sample, respectively. Represents the dot product.

8. The semantic-based meta-comparison sequence recommendation method according to claim 1, characterized in that: The specific steps of using the meta-update strategy to train and update the encoder and the regularizer to generate the sequence recommendation model include: Initialize the encoder and semantic regularizer, and back-propagate to update the encoder based on the recommendation loss, the contrastive loss of the first contrastive learning, and the recommendation loss of the second contrastive learning; The encoder parameters are fixed, the contrastive loss for the second contrastive learning is calculated based on the newly encoded sequence representation, and the semantic regularizer is updated by back-propagation.

9. An electronic device, characterized in that: include: processor; a memory having stored thereon one or more computer program instructions executed on the processor; Wherein, when the processor runs the computer instructions, it executes the semantic-based meta-comparison sequence recommendation method described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and when the computer instructions are executed by a processor, the steps of the semantic-based meta-comparison sequence recommendation method according to any one of claims 1 to 8 are executed.

Images