Explanatable recommendation system and method based on large language model
Through an interpretability recommendation system based on a large language model, using synergistic word segmentation and modules to generate real explanations, the problem of insufficient transparency of the existing recommendation system is solved and user trust and acceptance is enhanced.
Patent Information
- Application Number
- CN202510401701.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-01
- Publication Date
- 2025-07-04
AI Technical Summary
The lack of transparency and interpretability of existing recommendation systems leads to insufficient trust and acceptance of recommendation results.
An interpretability recommendation system based on a large language model is adopted to generate real explanations to improve system transparency through a collaborative relation word segmenter, collaborative information adapter, collaborative representation module, structured prompt module and training target module.
Enhance users' understanding and trust in the recommendation results, and support the optimization and improvement of the recommendation system.
Smart Images
Figure CN120256729A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of artificial intelligence, and specifically to an interpretable recommendation system and method based on a large language model. Background Technique
[0002] With the rapid development of artificial intelligence technology, the application of large language models in recommendation systems has become increasingly widespread. Recommendation systems have achieved great success in the Internet industry. Recommendation systems learn user preferences based on users' personal information and click-browse behaviors, etc., and thus recommend suitable objects for users according to their preferences.
[0003] However, existing recommendation systems are often regarded as "black boxes", and their decision-making processes lack transparency and interpretability, which limits users' trust and acceptance of recommendation results. Therefore, an interpretable recommendation system and method based on a large language model are proposed. Summary of the Invention
[0004] (1) Technical Problems to be Solved
[0005] Aiming at the deficiencies of the prior art, the present invention provides an interpretable recommendation system and method based on a large language model, mainly to solve the problem that existing recommendation systems are often regarded as "black boxes", and their decision-making processes lack transparency and interpretability, which limits users' trust and acceptance of recommendation results.
[0006] (2) Technical Solutions
[0007] To achieve the above object, the present invention provides the following technical solutions:
[0008] An interpretable recommendation system based on a large language model, including a system module assembly. The system module assembly includes a collaborative relationship tokenizer, a collaborative information adapter, a collaborative representation module, a model module, and a structured prompt module. The collaborative relationship tokenizer is used to obtain user representations and item representations. The collaborative relationship tokenizer is connected to the collaborative information adapter, and the collaborative information adapter is used to obtain the semantic representation between users and items. The model module constructs a large language model. The collaborative representation module is connected to the collaborative information adapter, and the collaborative representation module embeds the semantic representation into each layer of the large language model. The structured prompt module forms a structured prompt template, and the structured prompt module is connected to a training objective module. The training objective module trains the modified large language model. The training objective module is connected to a true explanation generation module, and the true explanation generation module is used to input the user's original evaluation of the item into the trained large language model to generate a true explanation.
[0009] Further, the user is represented as the encoding of user features or behaviors by the large language model, and the item is represented as the encoding of item features by the large language model.
[0010] Based on the foregoing solution, the semantic representation is the encoding of the deep semantic information of the input text by the large language model.
[0011] As a further solution of the present invention, the system module assembly includes a data collection module, and the data collection module is connected to the model module.
[0012] Further, the data collection module is connected to a processing module, and the processing module screens the collected data.
[0013] The present invention also proposes an interpretable recommendation method based on a large language model, including the following steps:
[0014] S1: Obtain data, use a collaborative relationship tokenizer to obtain user representations and item representations, and convert the features of users and items into a form recognizable by a computer through tokenization and vectorization techniques;
[0015] S2: Construct a semantic relationship, use a collaborative information adapter to obtain the semantic representation between users and items, and construct a semantic relationship by calculating the similarity or correlation between users and items;
[0016] S3: Representation injection, use a collaborative representation module to inject the semantic representation into each layer of the modified large language model, so that the model can learn the deep semantic relationship between users and items;
[0017] S4: Form a prompt template, use a structured prompt module to form a structured prompt template to provide clear guidance for training the target module;
[0018] S5: Generate recommendation results, use the training target module to train the modified large language model so that it can generate recommendation results according to the features of users and items;
[0019] S6: Generate a real explanation, use the real explanation generation module to input the user's original evaluation of the item into the trained large language model, and let it generate a real explanation to describe the reason for the recommendation result.
[0020] As a further solution of the present invention, the tokenization and vectorization techniques in S1 include a conversion module, and the conversion module is connected to an identification module.
[0021] Further, S4 includes a display module, and the prompt template is displayed through the display module.
[0022] On the basis of the foregoing solution, the S5 includes a recording module that records the generated recommendation results, and the recording module is connected to the display module.
[0023] (III) Beneficial Effects
[0024] Compared with the prior art, the present invention provides an interpretable recommendation system and method based on a large language model, having the following beneficial effects:
[0025] 1. The present invention obtains the semantic representation between users and items through a collaborative information adapter, and the collaborative representation module embeds the semantic representation into each layer of the large language model.
[0026] 2. The present invention trains the modified large language model through a training objective module, and inputs the original evaluation of the user on the item into the trained large language model through a true explanation generation module to generate a true explanation.
[0027] 3. The transparency of the recommendation system of the present invention enables users to better understand the decision-making process of the recommendation results, enhances the trust and acceptance of users in the recommendation system, and provides strong support for the optimization and improvement of the recommendation system. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 It is a schematic diagram of the module structure of an interpretable recommendation system based on a large language model proposed by the present invention.
[0029] Figure 2 It is a schematic diagram of the process structure of an interpretable recommendation method based on a large language model proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0031] Embodiment 1
[0032] Refer to Figure 1-2, An interpretable recommendation system based on a large language model, including a system module assembly. The system module assembly includes a collaborative relationship tokenizer, a collaborative information adapter, a collaborative representation module, a model module, and a structured prompt module. The collaborative relationship tokenizer is used to obtain user representation and item representation. The collaborative relationship tokenizer is connected to the collaborative information adapter, and the collaborative information adapter is used to obtain the semantic representation between the user and the item. The model module constructs a large language model. The collaborative representation module is connected to the collaborative information adapter, and the collaborative representation module embeds the semantic representation into each layer of the large language model. The structured prompt module forms a structured prompt template, and the structured prompt module is connected to a training objective module. The training objective module trains the modified large language model. The training objective module is connected to a real explanation generation module, and the real explanation generation module is used to input the user's original evaluation of the item into the trained large language model to generate a real explanation.
[0033] The user representation is the encoding of user features or behaviors by the large language model, the item representation is the encoding of item (such as goods, articles, etc.) features by the large language model, and the semantic representation is the deep semantic information encoding of the input text (such as words, sentences or paragraphs) by the large language model.
[0034] An interpretable recommendation method based on a large language model includes the following steps:
[0035] S1: Obtain data. Use the collaborative relationship tokenizer to obtain user representation and item representation, and convert the features of the user and the item into a form recognizable by a computer through tokenization and vectorization techniques;
[0036] S2: Construct a semantic relationship. Use the collaborative information adapter to obtain the semantic representation between the user and the item, and construct a semantic relationship by calculating the similarity or correlation between the user and the item;
[0037] S3: Representation injection. Use the collaborative representation module to inject the semantic representation into each layer of the modified large language model, so that the model can learn the deep semantic relationship between the user and the item;
[0038] S4: Form a prompt template. Use the structured prompt module to form a structured prompt template to provide clear guidance for the training objective module;
[0039] S5: Generate a recommendation result. Use the training objective module to train the modified large language model so that it can generate a recommendation result based on the features of the user and the item;
[0040] S6: Generate a real explanation. Use the real explanation generation module to input the user's original evaluation of the project into the trained large language model, which generates a real explanation to describe the reason for the recommendation result, the transparency of the recommendation system, enabling users to better understand the decision-making process of the recommendation result, enhancing users' trust and acceptance of the recommendation system, and providing strong support for the optimization and improvement of the recommendation system.
[0041] Embodiment 2
[0042] Refer to Figure 1-2 , an interpretable recommendation system based on a large language model, including a system module assembly. The system module assembly includes a collaborative relationship tokenizer, a collaborative information adapter, a collaborative representation module, a model module, and a structured prompt module. The collaborative relationship tokenizer is used to obtain user representation and item representation. The collaborative relationship tokenizer is connected to the collaborative information adapter, and the collaborative information adapter is used to obtain the semantic representation between the user and the item. The model module constructs a large language model. The collaborative representation module is connected to the collaborative information adapter, and the collaborative representation module embeds the semantic representation into each layer of the large language model. The structured prompt module forms a structured prompt template, and the structured prompt module is connected to a training objective module. The training objective module trains the modified large language model. The training objective module is connected to a real explanation generation module, and the real explanation generation module is used to input the user's original evaluation of the project into the trained large language model, which generates a real explanation.
[0043] The user representation is the encoding of user features or behaviors by the large language model, the item representation is the encoding of item (such as goods, articles, etc.) features by the large language model, the semantic representation is the deep semantic information encoding of the input text (such as words, sentences, or paragraphs) by the large language model. The system module assembly includes a data collection module. The data collection module is connected to the model module, and the data collection module is connected to a processing module. The processing module screens the collected data.
[0044] An interpretable recommendation method based on a large language model includes the following steps:
[0045] S1: Obtain data. Use the collaborative relationship tokenizer to obtain user representation and item representation, and convert the features of the user and the item into a form recognizable by a computer through tokenization and vectorization techniques;
[0046] S2: Construct a semantic relationship. Use the collaborative information adapter to obtain the semantic representation between the user and the item, and construct a semantic relationship by calculating the similarity or correlation between the user and the item;
[0047] S3: Representation injection. Use the collaborative representation module to inject the semantic representation into each layer of the modified large language model, enabling the model to learn the deep semantic relationship between the user and the item;
[0048] S4: Form a prompt template, and use the structured prompt module to form a structured prompt template to provide clear guidance for training the target module;
[0049] S5: Generate recommendation results. Use the training target module to train the modified large language model so that it can generate recommendation results based on the characteristics of users and projects;
[0050] S6: Generate real explanations. Use the real explanation generation module to input the user's original evaluation of the project into the trained large language model, and let it generate real explanations to describe the reasons for the recommendation results and the transparency of the recommendation system, enabling users to better understand the decision-making process of the recommendation results, enhancing users' trust and acceptance of the recommendation system, and providing strong support for the optimization and improvement of the recommendation system.
[0051] In particular, the tokenization and vectorization techniques in S1 include a conversion module, the conversion module is connected to an identification module, S4 includes a display module, the prompt template is displayed through the display module, S5 includes a recording module, the recording module records the generated recommendation results, and the recording module is connected to the display module.
[0052] In the description of this article, it should be noted that relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.
Claims
1. An interpretable recommendation system based on a large language model, including a system module assembly, characterized in that, The system module assembly includes a collaborative relationship tokenizer, a collaborative information adapter, a collaborative representation module, a model module, and a structured prompt module. The collaborative relationship tokenizer is used to obtain user representation and item representation. The collaborative relationship tokenizer is connected to the collaborative information adapter, and the collaborative information adapter is used to obtain the semantic representation between the user and the item. The model module constructs a large language model. The collaborative representation module is connected to the collaborative information adapter, and the collaborative representation module embeds the semantic representation into each layer of the large language model. The structured prompt module forms a structured prompt template, and the structured prompt module is connected to a training objective module. The training objective module trains the modified large language model. The training objective module is connected to a true explanation generation module, and the true explanation generation module is used to input the user's original evaluation of the item into the trained large language model to generate a true explanation.
2. The interpretable recommendation system based on a large language model according to claim 1, wherein The user representation is the encoding of user features or behaviors by the large language model, and the item representation is the encoding of item features by the large language model.
3. An interpretable recommendation system based on a large language model according to claim 1, characterized in that, The semantic representation is the deep semantic information encoding of the input text by the large language model.
4. An interpretable recommendation system based on a large language model according to claim 1, characterized in that, The system module assembly includes a data collection module, and the data collection module is connected to the model module.
5. An interpretable recommendation system based on a large language model according to claim 4, characterized in that, The data collection module is connected to a processing module, and the processing module screens the collected data.
6. An interpretable recommendation method based on a large language model, characterized in that, It includes the following steps: S1: Obtain data. Use the collaborative relationship tokenizer to obtain user representation and item representation, and convert the features of the user and the item into a form recognizable by a computer through tokenization and vectorization techniques. S2: Construct a semantic relationship. Use the collaborative information adapter to obtain the semantic representation between the user and the item, and construct a semantic relationship by calculating the similarity or correlation between the user and the item. S3: Representation injection. Use the collaborative representation module to inject the semantic representation into each layer of the modified large language model so that the model can learn the deep semantic relationship between the user and the item. S4: Form a prompt template. Use the structured prompt module to form a structured prompt template to provide clear guidance for the training objective module. S5: Generate a recommendation result. Use the training objective module to train the modified large language model so that it can generate a recommendation result based on the features of the user and the item. S6: Generate a true explanation. Use the true explanation generation module to input the user's original evaluation of the item into the trained large language model to generate a true explanation to describe the reason for the recommendation result.
7. An interpretable recommendation method based on a large language model according to claim 6, characterized in that, The tokenization and vectorization techniques in S1 include a conversion module, and the conversion module is connected to an identification module.
8. An interpretable recommendation method based on a large language model according to claim 6, characterized in that, In S4, it includes a display module, and the prompt template is displayed through the display module.
9. The interpretable recommendation method based on a large language model according to claim 8, wherein In S5, it includes a recording module, and the recording module records the generated recommendation result, and the recording module is connected to the display module.