Explanatable recommendation system, method and equipment based on large language model and medium
By introducing large language models and graph neural networks into the recommendation system to generate natural language interpretations, the problem of lack of transparency in the decision-making process of the existing recommendation system is solved, and user trust and recommendation strategy optimization capabilities are improved.
Patent Information
- Application Number
- CN202411060232.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-05
- Publication Date
- 2024-11-29
AI Technical Summary
The existing recommendation system lacks transparency and interpretability in the decision-making process, resulting in low user distrust and acceptance of recommendation results, and it is difficult for merchants to optimize and adjust recommendation strategies.
Adopting an interpretability recommendation system based on a large language model, natural language explanation is generated through collaborative relational word segmentation, collaborative information adapter, collaborative representation module, structured prompt module and training target module, and combining graph neural network and expert hybrid network to optimize Recommendation strategy.
It improves the transparency and accuracy of the recommendation system, enhances users' trust in recommendation results, and allows merchants to optimize and adjust recommendation strategies more effectively.
Smart Images

Figure CN119046529A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of artificial intelligence technology, and relates to an explainable recommendation system, method, device and medium, and in particular to an explainable recommendation system, method, device and medium based on a large language model. Background Art
[0002] Explainable recommendation refers to the technology that can provide corresponding explanations while providing recommendation results in the recommendation system. This technology enables users to understand why the recommendation system recommends a specific item, thereby increasing users' trust and satisfaction with the recommendation system and improving the transparency and usability of the recommendation system. In current recommendation systems, especially those based on machine learning and deep learning, the complexity of the recommendation process often makes the recommendation results difficult to explain. For example, although collaborative filtering, matrix decomposition, and deep learning models can provide accurate recommendations, the recommendation decision process is like a "black box" and users do not know how the recommendation is generated. The unexplainability in the above scenarios may cause users to distrust the recommendation results, especially when the recommendation results are far from user expectations.
[0003] Recommendation in the e-commerce field is a technology that analyzes user behavior and preferences to recommend products or services that users may be interested in. This technology can improve user experience and increase sales performance; at the same time, e-commerce recommendation technology needs to consider issues such as user privacy protection and algorithm transparency to ensure the security and ethics of technology application.
[0004] Large language model technology is an artificial intelligence technology that learns language rules and knowledge by training large-scale neural network models to achieve natural language understanding and generation. This technology usually uses deep learning algorithms, such as Transformer and RNN, to train models. Large language model technology has a wide range of applications, including text classification, machine translation, sentiment analysis, and intelligent customer service. Large language model technology requires a lot of computing resources and storage space during the training process, as well as a large amount of labeled data and high-quality unsupervised data. In order to improve the performance and effect of the model, researchers usually need to tune and optimize the model, such as adjusting the model structure, adding pre-training tasks, etc.
[0005] Although existing recommendation systems have made great progress in accuracy and performance, their decision-making processes often lack transparency and explainability, making it difficult for users to understand the reasons for the recommendation results, thus affecting users' trust and acceptance of the recommendation results.
[0006] In addition, the lack of explainability of the recommendation system also limits merchants' optimization and adjustment of recommendation strategies. They cannot understand the decision-making logic of the recommendation system, find and solve potential problems, and cannot make effective recommendation strategy adjustments based on user feedback and market demand.
[0007] Therefore, in view of the defects existing in the above-mentioned prior art, it is necessary to develop a new type of explainable recommendation system, method, device and medium. Summary of the invention
[0008] In order to overcome the defects of the prior art, the present invention proposes an explainable recommendation system, method, device and medium based on a large language model, aiming to solve the shortcomings of the existing e-commerce recommendation system in terms of explainability. By utilizing the large language model technology, it is possible to generate natural language explanations to describe the reasons for the recommendation results, so that users can better understand the decision-making process of the recommendation system. At the same time, the large language model technology can also continuously optimize the recommendation strategy by learning user feedback and behavior data, thereby improving the accuracy and performance of the recommendation system.
[0009] In order to achieve the above object, the present invention provides the following technical solutions:
[0010] An explainable recommendation system based on a large language model, characterized by comprising:
[0011] A collaborative relation tokenizer, which is used to represent users, items, and the interactive relationships between them to obtain user representation and item representation;
[0012] A collaborative information adapter, which is used to obtain a semantic representation between a user and a project based on the user representation and the project representation, thereby forming an adapted representation including the user representation, the project representation and the semantic representation;
[0013] A collaborative representation module, which is used to modify the projection functions of the key, query and value of each layer of the large language model so that the adapted representation can be injected into each layer of the modified large language model;
[0014] A structured prompt module, which is used to form a structured prompt template and embed the adaptation representation into the structured prompt template;
[0015] A training target module, which is used to input the structured prompt template into the modified large language model to train it;
[0016] The true explanation generation module is used to input the user's original evaluation of the project into the trained large language model, thereby generating the true explanation of the user's original evaluation of the project.
[0017] Preferably, the LightGCN model is used as the collaborative relationship segmenter, and the user representation and item representation obtained are respectively:
[0018]
[0019]
[0020] In the formula, is the user representation of user u output by the l+1th layer of the LightGCN model, N u is the neighbor set of user u, N i is the set of items interacted by user u, is the item representation of item i output by the lth layer of the LightGCN model, is the item representation of item i output by the l+1th layer of the LightGCN model, is the user representation of user u output by the lth layer of the LightGCN model.
[0021] Preferably, it further includes a collaborative relational word segmenter optimization module, which is used to use the Bayesian personalized ranking loss as a supervisory signal to guide the generation of user representation and item representation of the collaborative relational word segmenter, and the Bayesian personalized ranking loss is:
[0022]
[0023] In the formula, represents the predicted score between user u and item i, σ represents the Sigmoid function, represents the prediction score between user u and historical item j, j represents the historical items that user u has interacted with, and m represents the number of neighbors of user u.
[0024] Preferably, an expert hybrid network is used as the collaborative information adapter.
[0025] Preferably, the projection functions of the query, key, and value of each layer of the modified large language model are:
[0026] f q,k,v (x i )=W q,k,v ·a i
[0027] Where W q,k,v is the mapping matrix between query, key and value, a i For adaptation, x i It is a semantic mapping representation.
[0028] Preferably, the structured prompt template is:
[0029] P = [p {u} ,p {i} ,p {l} ]
[0030] Where P is the structured prompt template input into the modified large language model, p {u} is the user representation embedded in the structured prompt template, p {i} is the item representation embedded in the structured prompt template, p {l} is the semantic representation embedded in the structured prompt template.
[0031] Preferably, when training the modified large language model, negative log-likelihood is used as the training loss function, which is specifically:
[0032]
[0033] Where N is the number of interpretable segmentations, C i is the character count in each interpretable, y i,c and Represent the actual interpretable participle and the predicted interpretable participle respectively.
[0034] In addition, the present invention also provides an interpretable recommendation method based on a large language model, which is characterized in that it adopts the above-mentioned interpretable recommendation system based on a large language model and includes the following steps:
[0035] Represent users, items, and the interactions between them to obtain user representation and item representation;
[0036] Based on the user representation and the item representation, a semantic representation between the user and the item is obtained, thereby forming an adapted representation including the user representation, the item representation and the semantic representation;
[0037] Modify the projection functions of the key, query, and value of each layer of the large language model so that the adapted representation can be injected into each layer of the modified large language model;
[0038] forming a structured prompt template, and embedding the adapted representation into the structured prompt template;
[0039] Inputting the structured prompt template into the modified large language model to train it;
[0040] The user's original evaluation of the project is input into the trained large language model, which generates a true interpretation of the user's original evaluation of the project.
[0041] Preferably, it further comprises:
[0042] The Bayesian personalized ranking loss is adopted as the supervisory signal to guide the generation of user representation and item representation of the collaborative relation tokenizer.
[0043] Moreover, the present invention also provides an explainable recommendation device based on a large language model, characterized in that it includes:
[0044] one or more processors;
[0045] A memory for storing one or more programs;
[0046] When the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned explainable recommendation method based on the large language model.
[0047] Finally, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, the steps of the above-mentioned interpretable recommendation method based on a large language model are implemented.
[0048] Compared with the prior art, the interpretable recommendation system, method, device and medium based on a large language model of the present invention have one or more of the following beneficial technical effects:
[0049] 1. The present invention uses graph neural network as a word segmenter to encode complex user-item relationships into a latent embedding space, which allows the model to effectively capture user preferences and generate more accurate recommendations. The present invention propagates and aggregates information in the user-item interaction graph through an information passing mechanism, thereby learning the representation of user and item nodes.
[0050] 2. The present invention designs a lightweight adapter to integrate collaborative signals into the large language model. Since there may be differences in the semantic representation space between behavior-level collaborative information and user / item text semantics, the adapter aligns the behavior-aware collaborative signals with the text semantic space of LLM through a mixture of experts (MoE) method, effectively bridging the semantic gap between behavior-aware collaborative relationship words and text language words.
[0051] 3. The present invention introduces a collaborative instruction fine-tuning paradigm to align behavior-level information with language-level semantics, thereby incorporating user preferences into the knowledge of LLM. By adding special tags to input prompts and injecting adapted embeddings into reserved locations, the present invention is able to incorporate collaborative information into LLM. In addition, the present invention injects adapted embeddings into each layer of LLM to ensure that the model continuously accesses and integrates collaborative information throughout the entire structure. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] Figure 1It is a schematic diagram of the composition of the explainable recommendation system based on a large language model of the present invention.
[0053] Figure 2 It is a flowchart of the explainable recommendation method based on a large language model of the present invention. DETAILED DESCRIPTION
[0054] Before describing in detail any embodiment of the present invention, it should be understood that the present invention is not limited in its application to the construction and arrangement details of the components described below or illustrated in the following figures. The present invention can have other embodiments and can be practiced or carried out in various ways. In addition, it should be understood that the words and terms used here are for descriptive purposes and should not be considered restrictive. The use of "including" or "having" and its variations herein is intended to cover the items and their equivalents and additional items displayed below. Unless otherwise specified or limited, the terms "install", "connect", "support" and "couple" and their variations are widely used and cover direct installation and indirect installation, connection, support and connection. In addition, "connect" and "couple" are not limited to physical or mechanical connections or connections.
[0055] Furthermore, on the first aspect, in the disclosure of the present invention, the orientation or positional relationship indicated by terms such as "longitudinal", "transverse", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside" and "outside" are based on the orientation or positional relationship shown in the accompanying drawings, which are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore the above terms cannot be understood as limitations on the present invention; on the second aspect, the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the element may be multiple, and the term "one" cannot be understood as a limitation on the quantity.
[0056] The present invention relates to an explainable recommendation system, method, device and medium based on a large language model, which are specially designed to generate the explainability of user-item interactions in the e-commerce field, with the goal of revealing the basic reasons behind these interactions and clarifying the factors that affect user behavior. By unifying graph collaborative filtering and a large language model, it aims to provide an insightful explanation solution to help users understand the reasons why certain interactions occur and enhance the transparency of the recommendation process.
[0057] Figure 1 It is a schematic diagram of the composition of the explainable recommendation system based on a large language model of the present invention.
[0058] like Figure 1As shown, the interpretable recommendation system based on a large language model of the present invention includes:
[0059] 1. Collaborative relational tokenizer.
[0060] The coordination relationship word segmenter is used to represent users, items and the interaction relationship between them to obtain user representation and item representation.
[0061] Although natural language processing has its limitations, graph representation technology provides a powerful alternative. In order to efficiently capture the collaborative relationships between a large number of users and items and reflect their interaction patterns, this paper uses graph neural networks as tokenizers for representation and encodes high-order collaborative relationship information into the latent representation space, making it possible to effectively model complex user preferences.
[0062] Based on the message passing of the graph, this paper proposes a collaborative relationship segmenter, which aggregates and transmits information in the user-item interaction graph through the message passing mechanism, and promotes the learning of user and item node representations. This paper adopts the LightGCN model as a technology for efficient collaborative information aggregation, and the user representation and item representation obtained are:
[0063]
[0064]
[0065] In the formula, is the user representation of user u output by the l+1th layer of the LightGCN model, N u is the neighbor set of user u, N i is the set of items interacted by user u, is the item representation of item i output by the lth layer of the LightGCN model, is the item representation of item i output by the l+1th layer of the LightGCN model, is the user representation of user u output by the lth layer of the LightGCN model.
[0066] In order to optimize the collaborative relational tokenizer, the present invention further includes a collaborative relational tokenizer optimization module, which is used to use Bayesian personalized ranking loss as a supervisory signal to guide the generation of user representation and project representation of the collaborative relational tokenizer. That is, the present invention uses implicit feedback signals from user interactions and uses Bayesian personalized ranking (BPR) loss as a supervisory signal to guide the generation of user and project representations. Wherein, the Bayesian personalized ranking loss is:
[0067]
[0068] In the formula, represents the predicted score between user u and item i, σ represents the Sigmoid function, represents the prediction score between user u and historical item j, j represents the historical items that user u has interacted with, and m represents the number of neighbors of user u.
[0069] 2. Collaborative information adapter.
[0070] The collaborative information adapter is used to obtain the semantic representation between the user and the project based on the user representation and the project representation, thereby forming an adapted representation including the user representation, the project representation and the semantic representation.
[0071] In view of the possible differences in semantic representation space between behavior-level collaborative information and user- and project-related textual semantics, this paper proposes a lightweight and effective collaborative information adapter that is used to align these different modalities, enabling large language models to effectively utilize collaborative signals and textual semantics.
[0072] In order to bridge the semantic gap between the input of large language models (LLMs) and the collaborative relational word segmentation obtained by the present invention, and to enhance the generalization ability of large language models, the present invention adopts a mixture of experts (MoE) network as a collaborative information adapter. In this expert mixture network architecture, each expert network represents a different semantic dimension through a linear layer, and then these expert networks are integrated through a learnable gated routing mechanism. This enables the large language model to adaptively combine different semantic representations encoded by various experts, effectively bridging the gap between the perceived collaborative relational word segmentation and the text language word segmentation.
[0073] 3. Collaborative representation module.
[0074] The collaborative representation module is used to modify the projection functions of the keys, queries and values of each layer of the large language model so that the adapted representation can be injected into each layer of the modified large language model.
[0075] With the help of the proposed representation of the collaborative information adapter, the present invention can incorporate collaborative information into large language models. Space in the input prompt template is reserved by introducing special segmentations, and the adapted representations are injected into these reserved positions after the prompt is converted into segmentation representations. However, as the input length increases, each node representation is represented by only one segmentation in the input prompt, and the attention weight assigned to each representation inevitably decreases, resulting in a potential loss of collaborative information. To address the dilution of this effect, the present invention extends the injection of adapted representations beyond the initial input prompt. Specifically, the present invention incorporates them into the reserved positions of each layer of the large language model. To facilitate this process, the projection functions for queries, keys, and values of each layer of the modified large language model of the present invention are:
[0076] f q,k,v (x i )=W q,k,v ·a i
[0077] Where W q,k,v is the mapping matrix between query, key and value, a i For adaptation, x i It is a semantic mapping representation.
[0078] This ensures that the large language model continuously accesses and integrates collaborative information throughout the structure, not just at the input stage. By injecting the adapted representation into all layers of the LLMs, not only does it maintain a strong representation of the collaborative context, but it also enables the gradient flow to flow directly and efficiently back to the mixture of experts (MoE) network, enhancing the model's innovation. This innovative large language modeling and graph representation learning enables the model to leverage the deep contextual insights provided by the LLM while benefiting from the structural patterns recognized by the graph neural network.
[0079] 4. Structured prompt module.
[0080] The structured prompt module is used to form a structured prompt template and embed the adaptation representation into the structured prompt template.
[0081] The present invention uses a structured prompt template to integrate various data elements. This process includes segmenting the prompt and converting it into a representation space. In order to ensure that special segmented words in the prompt are recognized as unique entities, they are integrated into the LLM segmenter. These special segmented words are then replaced by their corresponding adapted representations and converted into transformed segmented representations. Specifically, the structured prompt template is:
[0082] P=[p {u} ,p {i} ,p {l} ]
[0083] Where P is the structured prompt template input into the modified large language model, p {u} is the user representation embedded in the structured prompt template, p {i} is the item representation embedded in the structured prompt template, p {l} is the semantic representation embedded in the structured prompt template.
[0084] Therefore, the present invention integrates user representation, item representation and semantic representation into a structured prompt template, which is convenient for training a large language model.
[0085] 5. Training target module.
[0086] The training target module is used to input the structured prompt template into the modified large language model to train it.
[0087] In order to improve the ability of large language models (LLMs) to generate contextually and grammatically coherent interpretations, the present invention aims to minimize the loss between the predicted probability of the next word segment and the actual next word segment in the sequence, using negative log-likelihood (NLL) as the training loss, calculated as follows:
[0088]
[0089] Here, N is the number of interpretable segmentations, C i is the character count in each interpretation, y i,c and Represent the actual and predicted interpretable participles, respectively.
[0090] To reduce training complexity, the present invention freezes all parameters within the LLMs, except for the parameters that interact with the graph neural network training process, and the only trainable parameters are the parameters within the expert mixture network.
[0091] 6. Real explanation generation module.
[0092] The true explanation generation module is used to input the user's original evaluation of the project into the trained large language model, so as to generate the true explanation of the user's original evaluation of the project.
[0093] After the modified large language model is trained, the collected original evaluations of the users on the items only need to be input into the trained large language model, and the model can generate the true interpretation of the original evaluations of the users on the items.
[0094] Figure 2 A flowchart of the explainable recommendation method based on a large language model of the present invention is shown.
[0095] like Figure 2As shown, the interpretable recommendation method based on a large language model of the present invention comprises the following steps:
[0096] 1. Represent users, items, and their interactions to obtain user representation and item representation.
[0097] In the present invention, the Bayesian personalized ranking loss can be used as a supervisory signal to guide the generation of user representation and item representation of the collaborative relational tokenizer.
[0098] 2. Based on the user representation and the project representation, a semantic representation between the user and the project is obtained, thereby forming an adapted representation including the user representation, the project representation and the semantic representation.
[0099] 3. Modify the projection functions of the keys, queries, and values of each layer of the large language model so that the adapted representation can be injected into each layer of the modified large language model.
[0100] 4. Form a structured prompt template, and embed the adaptation representation into the structured prompt template.
[0101] 5. Input the structured prompt template into the modified large language model to train it.
[0102] 6. Input the user's original evaluation of the project into the trained large language model, which generates a true interpretation of the user's original evaluation of the project.
[0103] In addition, the present invention also provides an explainable recommendation device based on a large language model, which includes: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the explainable recommendation method based on the large language model as described above.
[0104] Finally, the present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, the steps of the explainable recommendation method based on a large language model as described above are implemented.
[0105] The present invention proposes a new interpretable recommendation system in the e-commerce field based on a large language model, which seamlessly integrates the graph-based collaborative filtering paradigm and the capabilities of a large language model to generate comprehensive and profound explanations for the recommendation output. Moreover, the present invention effectively captures the high-order dependencies that form the basis of user preferences and item associations by leveraging the intrinsic collaborative relationships encoded in the user-item interaction graph. In addition, the present invention introduces a dedicated collaborative information adapter, which serves as a key bridge to establish a strong connection between collaborative signals and the rich text semantics encoded in the LLM. The interpretable recommendation system of the present invention has significant advantages. It not only enhances the interpretability of the recommendation process, but also ensures robustness, especially in challenging zero-shot recommendation scenarios, showing strong generalization capabilities.
[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Those skilled in the art can modify or replace the technical solution of the present invention according to the idea of the present invention without departing from the essence and scope of the technical solution of the present invention.
Claims
1. An explainable recommendation system based on a large language model, characterized in that: include: A collaborative relation tokenizer, which is used to represent users, items, and the interaction relationships between them to obtain user representation and item representation; A collaborative information adapter, which is used to obtain a semantic representation between a user and a project based on the user representation and the project representation, thereby forming an adapted representation including the user representation, the project representation and the semantic representation; A collaborative representation module, which is used to modify the projection functions of the key, query and value of each layer of the large language model so that the adapted representation can be injected into each layer of the modified large language model; A structured prompt module, which is used to form a structured prompt template and embed the adaptation representation into the structured prompt template; A training target module, which is used to input the structured prompt template into the modified large language model to train it; The true explanation generation module is used to input the user's original evaluation of the project into the trained large language model, thereby generating the true explanation of the user's original evaluation of the project.
2. The interpretable recommendation system based on a large language model according to claim 1, characterized in that: The LightGCN model is used as the collaborative relationship segmenter, and the user representation and item representation obtained are: In the formula, is the user representation of user u output by the l+1th layer of the LightGCN model, N u is the neighbor set of user u, N i is the set of items interacted by user u, is the item representation of item i output by the lth layer of the LightGCN model, is the item representation of item i output by the l+1th layer of the LightGCN model, is the user representation of user u output by the lth layer of the LightGCN model.
3. The interpretable recommendation system based on a large language model according to claim 2, characterized in that: Further comprising a collaborative relational word segmenter optimization module, which is used to use the Bayesian personalized ranking loss as a supervisory signal to guide the generation of user representation and item representation of the collaborative relational word segmenter, and the Bayesian personalized ranking loss is: In the formula, represents the predicted score between user u and item i, σ represents the Sigmoid function, represents the prediction score between user u and historical item j, j represents the historical items that user u has interacted with, and m represents the number of neighbors of user u.
4. The interpretable recommendation system based on a large language model according to claim 3, characterized in that: A mixture of experts network is used as the collaborative information adapter.
5. The interpretable recommendation system based on a large language model according to claim 4, characterized in that: The projection functions for queries, keys, and values at each layer of the modified large language model are: f q,k,v (x i )=W q,k,v ·a i Where W q,k,v is the mapping matrix between query, key and value, a i For adaptation, x i It is a semantic mapping representation.
6. The interpretable recommendation system based on a large language model according to claim 5, characterized in that: The structured prompt template is: P=[p{u},p{i},p{l}] Wherein, P is the structured prompt template input into the modified large language model, p{u} is the user representation embedded in the structured prompt template, p{i} is the item representation embedded in the structured prompt template, and p{l} is the semantic representation embedded in the structured prompt template.
7. The explainable recommendation system based on a large language model according to claim 6, characterized in that: When training the modified large language model, negative log-likelihood is used as the training loss function, which is specifically: Where N is the number of interpretable segmentations, C i is the character count in each interpretable, y i,c and Represent the actual interpretable participle and the predicted interpretable participle respectively.
8. An explainable recommendation method based on a large language model, characterized in that: It adopts the explainable recommendation system based on a large language model as described in any one of claims 1 to 7, and includes the following steps: Represent users, items, and the interactions between them to obtain user representation and item representation; Based on the user representation and the item representation, a semantic representation between the user and the item is obtained, thereby forming an adapted representation including the user representation, the item representation and the semantic representation; Modify the projection functions of the key, query, and value of each layer of the large language model so that the adapted representation can be injected into each layer of the modified large language model; forming a structured prompt template, and embedding the adapted representation into the structured prompt template; Inputting the structured prompt template into the modified large language model to train it; The user's original evaluation of the project is input into the trained large language model, which generates a true interpretation of the user's original evaluation of the project.
9. The interpretable recommendation method based on a large language model according to claim 8, characterized in that: Further including: The Bayesian personalized ranking loss is adopted as the supervisory signal to guide the generation of user representation and item representation of the collaborative relation tokenizer.
10. An explainable recommendation device based on a large language model, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the explainable recommendation method based on a large language model as described in claim 8 or 9.
11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the explainable recommendation method based on a large language model as described in claim 8 or 9 are implemented.
Citation Information
Cited By
Data processing method and device, information recommendation system, electronic equipment and medium
CN120011550A
Data processing method and apparatus, information recommendation system, electronic device, and medium
CN120011550B
Inplanatable recommendation method and system based on collaborative optimization of double large language models
CN121388294A
User behavior sequence-based interpretable recommendation generation method and recommendation system
CN121481679A
