Method and system for generating experience feedback recommendation based on large language model

Abstract

The invention belongs to the technical field of artificial intelligence recommendation, and particularly relates to an experience feedback recommendation generation method and system based on a large language model. Comprising a data collection and preprocessing module, a pre-training large language model selection module, a large language model feature extraction module, a model fine tuning module, an experience feedback recommendation generation module and a feedback and improvement module. The method has the beneficial effects that a large language model technology is applied to recommendation of experience feedback, and natural language text information generated in the interaction process of the user and the system is fully utilized, so that the user demand and intention are more accurately captured; and through semantic matching and context understanding, the recommendation accuracy and individuation degree are improved, so that the recommendation is more accurate and meets the actual requirements of the user.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence recommendation, and specifically relates to a method and system for generating experience feedback recommendations based on a large language model. Background Art

[0002] Experience feedback is an important feedback mechanism in the field of nuclear safety, and is also a key mechanism to ensure the quality of nuclear power plant equipment, prevent problems from recurring, and implement nuclear safety responsibilities. As a solid barrier to the safety of nuclear power plants, it is essentially a process of reusing solutions to various problems in nuclear power plants (including good practices). This is a management activity with clear goals, and its successful implementation depends on the effective collection and use of various "experiences" to ensure the safe operation and continuous improvement of nuclear power plants.

[0003] With the rapid development of information technology, recommendation systems have become an indispensable part of various online services. Traditional recommendation systems mainly rely on collaborative filtering, content filtering and other methods, and most of them make recommendations based on structured data such as user behavior and preferences, often ignoring the natural language text information generated by users in the process of interacting with the system. This information contains rich user intentions, needs and emotions, which is of great significance for improving the accuracy and personalization level of recommendation systems. At the same time, traditional recommendation methods are often limited by data sparsity, cold start problems and the degree of personalization of recommendation results.

[0004] It can be seen that traditional recommendation algorithms rely too much on structured data such as user behavior. Structured data is often sparse and difficult to fully reflect user needs and preferences. At the same time, due to the lack of effective extraction of natural language text information, although traditional recommendation systems can give accurate recommendation results, they lack explanations of recommendation logic, making it difficult for users to truly understand and trust them, and the level of personalization of recommendations needs to be improved. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for generating experience feedback recommendations based on a large language model to solve the problems of sparse data, poor interpretability, lack of feedback mechanism, cold start and low personalization in traditional recommendation systems.

[0006] The technical solution of the present invention is as follows: an experience feedback recommendation generation system based on a large language model, comprising a data collection and preprocessing module, a pre-trained large language model selection module, a large language model feature extraction module, a model fine-tuning module, an experience feedback recommendation generation module and a feedback and improvement module.

[0007] The data collection and preprocessing module mines and collects the interaction data between users and the system, including natural language text information of users' search queries, comments, reports, and feedback, and collects historical data of nuclear power plant experience feedback, performs preprocessing operations such as data cleaning, feature engineering, and data conversion, and converts the original data into a format that can be processed by a large language model.

[0008] The pre-trained large language model selection module selects an open source LLM model structure according to a specific task.

[0009] The large language model feature extraction module uses the selected large language model to process user data, including feature extraction and feature fusion. It analyzes the user's natural language text based on the user's historical behavior and personal information, extracts the user's interests, preferences and needs, and obtains the user's potential interests and preferences for specific content through model reasoning and prediction.

[0010] The model fine-tuning module uses the selected pre-trained model for fine-tuning and uses the historical experience feedback data of the nuclear power plant for model fine-tuning.

[0011] The experience feedback recommendation generation module matches the user's interests and needs with the recommended content library based on the output of the large language model processing module, and pushes the finally generated experience feedback recommendation results to the user.

[0012] A method for generating experience feedback recommendations based on a large language model, comprising the following steps:

[0013] Step 1: Data collection and preprocessing;

[0014] Step 2: Pre-trained large language model selection;

[0015] Step 3: The large language model extracts data features;

[0016] Step 4: Model fine-tuning;

[0017] Step 5: Experience feedback recommendation generation;

[0018] Step 6: Feedback and Improvement.

[0019] The step 1 collects user behavior data and historical experience feedback data from the nuclear power plant experience feedback system, and the acquired data is converted into a data format for large language model processing after preprocessing of data cleaning, normalization and standardization, word segmentation, stop word removal, and stem extraction.

[0020] In the step 3, the user data is processed using the large language model selected in step 2, and the deep structure of the large language model is used to extract features related to user interests, preferences and needs from the text vector, and the text data is converted into a vector representation. The feature information from different sources is integrated to form a feature vector that fully reflects the user's needs. In the large language model processing module, first, feature extraction performs deep processing on user behavior data by using the pre-trained weights of the large language model, converts the text into a high-dimensional vector, encodes each word or phrase in the text, generates a corresponding word vector, and gradually captures the deep features of the text through the multi-layer structure of the model.

[0021] After fine-tuning the model in step 5, the user data processed by the large language model is input into the large language model, and the user's potential needs and interests are captured through analysis and comparison, thereby generating a personalized experience feedback recommendation list.

[0022] In step 6, the user provides feedback on the recommendation results, including direct evaluation and behavioral data, and the recommendation algorithm is iteratively optimized by utilizing these data, including adjusting the recommendation strategy and optimizing the model parameters.

[0023] The beneficial effects of the present invention are as follows: the present invention applies the large language model technology to the recommendation of experience feedback, making full use of the natural language text information generated by the user in the process of interacting with the system, so as to more accurately capture the user's needs and intentions, and improve the accuracy and personalization of the recommendation through semantic matching and context understanding, so that the recommendation is more accurate and meets the actual needs of the user. Through the deep mining and utilization of experience feedback data, the reuse rate of the nuclear power plant experience feedback system is improved, and intelligent support is provided for the decision-making of the enterprise; the present invention realizes intelligent recommendation of experience feedback based on the large language model, which can process and understand diverse input data, including text, images, etc. These unstructured data enable the recommendation system to combine more diverse information for recommendation, so as to better solve the cold start and data sparsity problems in the traditional recommendation model. Utilizing the generation capability of the large language model, the recommendation system can also provide users with more informative and personalized recommendation explanations, enhancing users' understanding and acceptance of the recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A flowchart of a method for generating experience feedback recommendations based on a large language model provided by the present invention;

[0025] Figure 2 This is an architecture diagram of an experience feedback recommendation generation system based on a large language model provided by the present invention;

[0026] Figure 3A recommendation schematic diagram of an experience feedback recommendation generation method based on a large language model provided by the present invention. DETAILED DESCRIPTION

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0028] like Figure 2 As shown, an experience feedback recommendation generation system based on a large language model includes: a data collection and preprocessing module, a pre-trained large language model selection module, a large language model feature extraction module, a model fine-tuning module, an experience feedback recommendation generation module and a feedback and improvement module;

[0029] The data collection and preprocessing module mines and collects the interaction data between users and the system, including natural language text information such as users' search queries, comments, reports, feedback, etc., and collects historical data of experience feedback from nuclear power plants, performs preprocessing operations such as data cleaning, feature engineering, and data conversion, and converts the original data into a format suitable for large language model processing.

[0030] The pre-trained large language model selection module selects the appropriate open source LLM model structure according to the specific task, such as Transformer, LSTM, GRU, etc. When designing, factors such as the model size (number of parameters), number of layers, and attention mechanism need to be considered.

[0031] The large language model feature extraction module uses the selected large language model to process user data, including feature extraction and feature fusion. Based on the user's historical behavior and personal information, the large language model can deeply understand and analyze the user's natural language text and extract the user's interests, preferences and needs. Through the model's reasoning and prediction capabilities, we can obtain the user's potential interests and preferences for specific content.

[0032] The model fine-tuning module uses the selected pre-trained model for fine-tuning to adapt to specific recommendation tasks. Fine-tuning helps improve the performance of the model in specific fields or tasks. Here, the historical experience feedback data of nuclear power plants is used for model fine-tuning to improve the performance of the model on experience feedback related tasks.

[0033] The experience feedback recommendation generation module is based on the output of the large language model processing module. The recommendation generation module matches the user's interests and needs with the recommended content library and pushes the final generated experience feedback recommendation results to the user. This module not only considers the user's historical behavior and preferences, but also combines the large language model's in-depth understanding of the user's current intentions to generate more accurate and personalized recommendation results for the user.

[0034] After the feedback and improvement module generates the recommendation results, the system will display the recommended experience feedback content to the user. The user can interact with the system by clicking, browsing, and providing feedback. The system will adjust the recommendation strategy in real time based on the user's feedback and continuously optimize the recommendation results. At the same time, user feedback will also be used to further train and optimize the large language model to improve the model's ability to understand user intent.

[0035] like Figure 1 As shown, the present invention provides an experience feedback recommendation generation method based on a large language model, comprising the following steps:

[0036] Step 1: Data collection and preprocessing

[0037] Collect user behavior data and historical experience feedback data from the nuclear power plant experience feedback system. After data cleaning, normalization and standardization, word segmentation, stop word removal, stem extraction and other pre-processing steps, the acquired data is converted into a data format that is easy for large language models to process. Its role is: data collection and pre-processing ensure the accuracy, completeness and consistency of data. Through this process, it can clean missing values ​​and outliers, filter out invalid or redundant information, and provide high-quality data support for subsequent model training and decision support, thereby enhancing the reliability and effectiveness of data analysis.

[0038] Specifically, the collected user behavior data includes the user's browsing history, search query history, comments, reporting, feedback and other behavior data, and the data is stored in the database to store various types of user behavior information in real time. Collect existing nuclear power historical experience data, clean the data, and obtain data including but not limited to "number", "topic", "state description", "involved system", "event code", "action suggestion", "potential consequences", "event level" and "responsible department" and other field information; organize the obtained data into long text information in a specified format, complete the training data preprocessing operation, and use the organized data set as the training corpus for subsequent experiments.

[0039] Step 2: Pre-trained large language model selection

[0040] When choosing a pre-trained large language model, it is necessary to select a suitable model based on the requirements of the application scenario and resource constraints. Since the recommendation algorithm is to be introduced into the nuclear power plant experience feedback system, the selection of the algorithm should consider open source, easy deployment, model performance, model scale and available resources. You can consider the ChatGLM-6B large model, the Wenxin Yiyan large model (ERNIE-Bot) and the Tongyi Qianwen large model (Qwen-72B). ChatGLM-6B is an open source, Chinese and English bilingual question and answer dialogue language model, and is optimized for Chinese. Wenxin Yiyan large model (ERNIE-Bot) is a large-scale language model based on deep learning developed by Baidu Research Institute. The model covers a large amount of Chinese data, can learn rich language knowledge and semantic understanding ability, and has a good effect when processing Chinese questions and answers. The Qwen-72B model is the result of training on 3 trillion tokens of data. This dataset contains corpora in multiple languages ​​such as Chinese and English, as well as code, mathematics and other data.

[0041] Step 3: Large language model extracts data features

[0042] Use the large language model selected in step 2 to process user data, and use the deep structure of large language models (such as the GPT series, BERT, etc.) to extract features related to user interests, preferences, and needs from text vectors, and convert text data into vector representations. Next, we use feature fusion technology to integrate feature information from different sources to form a feature vector that fully reflects user needs. Through deep learning technology, automatic feature selection and fusion are achieved, providing strong support for subsequent recommendation generation.

[0043] In the large language model processing module, feature extraction and feature fusion are key steps. First, feature extraction deeply processes user behavior data by utilizing the pre-trained weights of large language models (such as BERT, GPT, etc.). The model converts text into high-dimensional vectors that contain the semantic and contextual information of the text. Specifically, the model encodes each word or phrase in the text, generates the corresponding word vector, and gradually captures the deep features of the text through the multi-layer structure of the model.

[0044] Next, feature fusion integrates feature information from different sources. Since large language models process data of multiple modalities (such as text, images, etc.), it is necessary to effectively fuse the features of these different modalities. A common method is to use an attention mechanism, which can weight different features according to their importance to obtain a unified feature representation. This fused feature representation can more comprehensively reflect the needs and interests of users, providing strong support for subsequent recommendation algorithms.

[0045] Step 4: Model fine-tuning

[0046] In the process of fine-tuning the large language model, it is necessary to combine the user behavior data collected in step 1 and the historical experience feedback data of the nuclear power plant. The user behavior data can reflect the user's preferences and behavioral habits, and the historical experience feedback data provides richer semantic information for the recommendation. When implementing fine-tuning, these two types of data need to be integrated and analyzed in order to optimize the output results of the model. Specifically, by building a suitable data pipeline, the collected data can be input into the model training process, and the pre-trained large language model can be iteratively trained. By adjusting hyperparameters such as the learning rate and the number of iterations, adding relevant features, and using appropriate optimization algorithms, the model's prediction accuracy and personalized recommendation capabilities can be improved.

[0047] Step 5: Experience feedback recommendation generation

[0048] Step 4: After fine-tuning the model, the large language model can better handle experience feedback-related tasks. The user data processed by the large language model is input into the large language model. The model can deeply understand user behavior and historical experience feedback, and combine it with the company's internal experience database to filter out the most relevant and valuable experience feedback for users. Through intelligent analysis and comparison, the algorithm can accurately capture the user's potential needs and interests, thereby generating a personalized experience feedback recommendation list. These recommendations not only take into account the user's immediate needs, but also incorporate long-term behavioral trends to ensure the accuracy and effectiveness of the recommendations.

[0049] Step 6: Feedback and Improvement

[0050] Users can provide feedback on recommendation results, including direct evaluation and behavioral data, to comprehensively evaluate the recommendation effect. By using this data to iteratively optimize the recommendation algorithm, such as adjusting the recommendation strategy and optimizing model parameters, the accuracy and personalization of the recommendation can be improved. In addition, we also incorporate user feedback into the continuous training of the large language model so that the model can continuously learn and adapt to changes in user needs. Its role is: through this closed-loop feedback and improvement mechanism, the performance of the recommendation system can be continuously improved, providing users with more accurate and valuable experience feedback recommendations.

[0051] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. An experience feedback recommendation generation system based on a large language model, characterized by: It includes data collection and preprocessing module, pre-trained large language model selection module, large language model feature extraction module, model fine-tuning module, experience feedback recommendation generation module and feedback and improvement module.

2. The experience feedback recommendation generation system based on a large language model as claimed in claim 1, characterized in that: The data collection and preprocessing module mines and collects the interaction data between users and the system, including natural language text information of users' search queries, comments, reports, and feedback, and collects historical data of nuclear power plant experience feedback, performs preprocessing operations such as data cleaning, feature engineering, and data conversion, and converts the original data into a format for large language model processing; The pre-trained large language model selection module selects an open source LLM model structure according to a specific task.

3. The experience feedback recommendation generation system based on a large language model as claimed in claim 1, characterized in that: The large language model feature extraction module uses the selected large language model to process user data, including feature extraction and feature fusion. It analyzes the user's natural language text based on the user's historical behavior and personal information, extracts the user's interests, preferences and needs, and obtains the user's potential interests and preferences for specific content through model reasoning and prediction.

4. The experience feedback recommendation generation system based on a large language model as claimed in claim 1, characterized in that: The model fine-tuning module uses the selected pre-trained model for fine-tuning and uses the historical experience feedback data of the nuclear power plant for model fine-tuning.

5. The experience feedback recommendation generation system based on a large language model as claimed in claim 1, characterized in that: The experience feedback recommendation generation module matches the user's interests and needs with the recommended content library based on the output of the large language model processing module, and pushes the finally generated experience feedback recommendation results to the user.

6. A method for generating experience feedback recommendations based on a large language model, characterized in that: The following steps are involved: Step 1: Data collection and preprocessing; Step 2: Pre-trained large language model selection; Step 3: The large language model extracts data features; Step 4: Model fine-tuning; Step 5: Experience feedback recommendation generation; Step 6: Feedback and Improvement.

7. The experience feedback recommendation generation method based on a large language model as claimed in claim 6, characterized in that: The step 1 collects user behavior data and historical experience feedback data from the nuclear power plant experience feedback system, and the acquired data is converted into a data format for large language model processing after preprocessing of data cleaning, normalization and standardization, word segmentation, stop word removal, and stem extraction.

8. The experience feedback recommendation generation method based on a large language model as claimed in claim 6, characterized in that: In the step 3, the user data is processed using the large language model selected in step 2, and the deep structure of the large language model is used to extract features related to user interests, preferences and needs from the text vector, and the text data is converted into a vector representation. The feature information from different sources is integrated to form a feature vector that fully reflects the user's needs. In the large language model processing module, first, feature extraction performs deep processing on user behavior data by using the pre-trained weights of the large language model, converts the text into a high-dimensional vector, encodes each word or phrase in the text, generates a corresponding word vector, and gradually captures the deep features of the text through the multi-layer structure of the model.

9. The experience feedback recommendation generation method based on a large language model according to claim 6, characterized in that: After fine-tuning the model in step 5, the user data processed by the large language model is input into the large language model, and the user's potential needs and interests are captured through analysis and comparison, thereby generating a personalized experience feedback recommendation list.

10. The experience feedback recommendation generation method based on a large language model according to claim 6, characterized in that: In step 6, the user provides feedback on the recommendation results, including direct evaluation and behavioral data, and the recommendation algorithm is iteratively optimized by utilizing these data, including adjusting the recommendation strategy and optimizing the model parameters.

Images