A method for intelligent recommendation state report based on BERT algorithm in Xinxing environment

Abstract

The application belongs to the technical field of information processing, and particularly relates to a method for intelligently recommending a state report based on a BERT algorithm in a Xinhua environment. The method comprises the following steps: step 1, obtaining a state report metadata set; step 2, data acquisition and preprocessing; step 3, BERT model training and optimization; step 4, feature extraction and representation learning; and step 5, an intelligent recommendation engine. The application has the beneficial effect of improving recommendation accuracy: through semantic understanding and multi-source data fusion of the BERT algorithm, the user's interest and demand can be more accurately captured, and the accuracy and quality of the state report recommendation are improved.

Description

Technical Field

[0001] This invention belongs to the field of information processing technology, specifically relating to a method for intelligent recommendation status reporting based on the BERT algorithm in the context of domestic information technology innovation. Background Technology

[0002] With the rapid development of information technology, intelligent recommendation systems have been widely applied in various fields. Traditional recommendation algorithms are mainly based on collaborative filtering and content filtering, but these methods have certain limitations when processing complex semantic information and long text data. BERT (Bidirectional Encoder Representations from Transformers), as an advanced natural language processing technology, can better understand the semantic information of text, providing new ideas and methods for intelligent recommendation.

[0003] Experience feedback is a core mechanism for safety culture and continuous improvement in the nuclear power industry. By systematically collecting, analyzing, sharing, and applying experiences (including incidents, anomalies, and best practices) from the operation, maintenance, and design of nuclear power plants both domestically and internationally, it has significantly improved the safety, reliability, and economy of nuclear power.

[0004] In experience feedback activities, the data used to describe performance deviations is called condition report data, and the system that carries this data is called a condition report system. This system is widely used in nuclear power plants, with tens of millions of data entries.

[0005] Currently, most intelligent recommendation systems rely on foreign technologies and products, posing risks such as data security and privacy breaches. In the context of domestic IT innovation, fully leveraging the data value of status reports to achieve independently controllable intelligent recommendation technology can, while ensuring national security, improve the efficiency of business personnel in learning from experience through accurate and intelligent recommendations of status reports, thereby reducing business training costs and the number of incidents. Summary of the Invention

[0006] The purpose of this invention is to provide a method for intelligent recommendation status reporting based on the BERT algorithm in the context of information technology innovation. By combining user behavior profiles, the method can achieve fast, accurate, and personalized push of status reports, improve the efficiency and quality of information transmission, and provide strong support for the development of the information technology innovation industry.

[0007] The technical solution of this invention is as follows: A method for intelligent recommendation status reporting based on the BERT algorithm in a domestic information technology innovation environment, comprising the following steps: Step 1: Obtain the status report metadata set; Step 2: Data Acquisition and Preprocessing; Step 3: BERT model training and optimization; Step 4: Feature extraction and representation learning; Step 5: Intelligent Recommendation Engine.

[0008] Step 2 includes: Step 21: Collect user behavior data, status report data, and related text data from multiple data sources in the domestic IT innovation environment.

[0009] Step 22: Clean, denoise, and standardize the collected data to improve data quality.

[0010] Step 23: Perform natural language processing operations such as word segmentation and part-of-speech tagging on the text data to construct a data format suitable for the BERT algorithm input.

[0011] The user behavior data in step 21 includes filling in information, browsing, rating, liking, and saving.

[0012] Step 3 includes: Step 31: Build a BERT model training platform in the domestic IT innovation environment, and train the BERT model using the preprocessed dataset; Step 32: Employ gradient descent and attention mechanisms to improve the performance and accuracy of the BERT model in the context of information technology innovation; Step 33: Evaluate and validate the trained BERT model to ensure its stability and reliability in the information technology innovation environment.

[0013] Step 4 includes: Step 41: Use the trained BERT model to extract features from the status report text and convert the text into a low-dimensional vector representation; Step 42: Learn the feature representation of the status report, mine the key information and semantic features, and provide a basis for subsequent recommendations.

[0014] Step 5 includes: Step 51: Define a weight table for user behavior in the status reporting system, and configure the weights for user actions such as filling in information, browsing, commenting, liking, and collecting. Step 52: Combining the user behavior weight table and user behavior information, select N sets of status reports with the highest interaction degree from all the status report data that the user has interacted with: SET1: {CR1,…,CR N This represents the user's interest value; Step 53: Based on the feature representations of the N status report data SET1 with the highest interaction weight scores selected in the previous step, apply the semantic matching method of the BERT algorithm to match the status report scores of the top M similarity scores to map MAP1:{{CR1,0.9},…,{CR M}}, and remove the status report data that has been interacted with by the user, to obtain a new ordered set of status report data MAP1':{{CR1,Score1},…,{CR M’， Score M’}}, where Score = Similarity * Interaction Weight Score; Step 54: Based on the user's department information, their job information in various systems of the company, and the time decay factor, adjust the sorting of status report scores in MAP1' to generate a personalized intelligent recommendation set of status report data for the user; Step 55: Adjust and optimize the recommendation results based on users' real-time feedback and behavioral data to improve the accuracy and satisfaction of the recommendations.

[0015] The beneficial effects of this invention are as follows: Improve recommendation accuracy: Through semantic understanding and multi-source data fusion using the BERT algorithm, user interests and needs can be captured more accurately, improving the accuracy and quality of status report recommendations.

[0016] Ensuring data security: The implementation of independently controllable intelligent recommendation technology in the context of information technology innovation avoids dependence on foreign technologies and reduces the risk of data security and privacy leaks.

[0017] Improved system performance: By adopting domestically produced hardware and software platforms and optimized algorithms, the system's operating speed and efficiency have been improved, enabling it to meet the needs of large-scale data processing and real-time recommendations.

[0018] Promoting industrial development: The application of this technology will help promote the development of the information technology innovation industry in the field of intelligent recommendation and enhance the competitiveness of domestic enterprises in the field of artificial intelligence. Attached Figure Description

[0019] Figure 1 Metadata for status reports; Figure 2 Flowchart for intelligent recommendation; Figure 3 Flowchart of the BERT model platform setup process; Figure 4 This describes the processing procedure for the Hugging Face Datasets library. Detailed Implementation

[0020] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] A method for intelligent recommendation status reporting based on the BERT algorithm in a domestic information technology application environment includes the following steps: Step 1: Obtain all status report metadata, user behavior data, user role information, etc. from the standardized status reporting systems of each nuclear power plant; The data source is the data generated by the status reporting system of each nuclear power plant, which exists in the database and can be fully obtained through the data platform.

[0022] Status report metadata mainly includes the status report source, status report subject, status report description, problem area, relevant expertise, and problem consequences.

[0023] User behavior data mainly includes records such as user status report submissions, browsing, ratings, likes, and favorites.

[0024] Step 2: Apply for and set up a domestic IT innovation environment (Alibaba Cloud V3 operating system, PolarDB-X distributed database), and migrate the acquired data to the domestic IT innovation environment.

[0025] Step 3: Set up the BERT model platform; Step 31: Prepare the NVIDIA GPU and install the underlying environment such as CUDA and PyTorch; Step 32: Use the Hugging Face Datasets library to clean and label the acquired data, and build a standard data format suitable for the input of the BERT algorithm.

[0026] Step 33: Load the pre-trained BERT, configure the parameters, fine-tune the BERT model parameters using the Trainer API, extract features from the status report text, and convert the text into a 768-dimensional vector representation. Step 34: Save the trained model and deploy it as an API service for applications to call.

[0027] Step 4: Intelligent Recommendation Engine Step 41: Define a weight table for user behavior in the status reporting system, and configure weights for user behaviors such as filling in information, browsing, rating, liking, and collecting.

[0028] Step 52: Quantify user behavior into behavior weight values ​​(e.g., 5 points for filling in information, 1 point for browsing, etc.), generate a user behavior weight table, and combine user behavior information with the algorithm principle of content filtering to select N status report sets SET1: {CR1,…,CRN} with the highest interaction and the user's most interest from all status report data that the user has interacted with. This set represents the user's status report data interest set.

[0029] CR stands for Condition Report.

[0030] Step 53: Based on the feature representation of the N status report data SET1 (the user's interest data set) with the highest interaction weight scores selected in the previous step, apply the semantic matching method of the BERT algorithm to match the status report scores of the top M status reports in SET1 with CR similarity scores, mapping MAP1:{{CR1, 0.9},…,{CR M The status report data that has been interacted with by the user is removed, resulting in a new ordered set of status report data MAP1':{{CR1,Score1},…,{CRM',ScoreM'}}, where Score = similarity * interaction weight score, representing the user's interest score in the status report data.

[0031] CR1 and CRM represent a specific status report data; Step 54: Based on the user's department information, their job information in various systems within the company, and the time decay factor, adjust the sorting of status report scores in MAP1' to generate a personalized intelligent recommendation set of status report data for the user.

[0032] Step 55: Based on real-time user feedback and behavioral data, continuously adjust the parameters of the BERT training model and the user behavior weight table parameters to improve the accuracy and satisfaction of recommendations.

[0033] Example: (I) Data Acquisition and Preprocessing 1. Determine the data source User behavior data and status report data are collected from the status reporting systems of each power plant.

[0034] 2. Apply for and build the information technology innovation environment resources.

[0035] 3. Use the Hugging Face Datasets library to clean and label the collected data, and standardize the data to convert data of different formats and units into a format suitable for model training and processing. For example, perform unified encoding processing on text data.

[0036] (II) BERT Model Training and Optimization 1. Set up a training platform To build a BERT model training platform in a domestic IT environment, select a deep learning framework, such as [specific framework name], and configure the corresponding computing resources, such as a GPU cluster of [specific chip name].

[0037] 2. Model Training The preprocessed dataset is divided into training, validation, and test sets. The BERT model is trained using the training set. Appropriate training parameters are set, such as learning rate, batch size, and number of training epochs, and the gradient descent algorithm is used for model optimization.

[0038] 3. Model Evaluation and Validation During training, the model is periodically evaluated using a validation set, and metrics such as accuracy, recall, and F1 score are calculated. Based on the evaluation results, the model's parameters and structure are adjusted until optimal performance is achieved. Finally, the model is validated using a test set to ensure its generalization ability on unknown data.

[0039] (III) Intelligent Recommendation Engine 1. Building a recommendation model By combining feature representations from user behavior data and status reports, an intelligent recommendation model is constructed: Define a weight table for user behavior in the status reporting system, and configure weights for user behaviors such as filling in information, browsing, rating, liking, and saving.

[0040] By combining user behavior weight tables and user behavior information, and comprehensively applying traditional recommendation algorithms such as collaborative filtering and content filtering, N sets of state reports with the highest interaction degree, SET1, are selected from all state report data that the user has interacted with: {CR1,…,CR...} N This represents the user's interest value.

[0041] Based on the feature representation of the N status report data SET1 with the highest interaction weight scores selected in the previous step, the semantic matching method of the BERT algorithm is applied to match the top M status report scores with similarity scores, mapping MAP1:{{CR1,0.9},…,{CR M}}, and remove the status report data that has been interacted with by the user, to obtain a new ordered set of status report data MAP1':{{CR1,Score1},…,{CR M’， Score M’}}, where Score = Similarity * Interaction Weight Score.

[0042] Based on the user's department information, their job information in various company systems, and the time decay factor, the status report scores in MAP1' are sorted to generate a personalized intelligent recommendation set of status report data for the user.

[0043] Based on real-time user feedback and behavioral data, we continuously adjust and optimize recommendation results to improve the accuracy and satisfaction of recommendations.

[0044] 2. Real-time feedback and optimization Based on real-time user feedback and behavioral data, such as clicks, browsing, and favorites, the recommendation results are continuously adjusted and optimized. User feedback data can be used as new training data to further train and optimize the BERT model and recommendation algorithm, improving the accuracy and satisfaction of recommendations.

Claims

1. A method for intelligent recommendation status reporting based on the BERT algorithm in a domestic information technology innovation environment, characterized in that, Includes the following steps: Step 1: Obtain the status report metadata set; Step 2: Data Acquisition and Preprocessing; Step 3: BERT model training and optimization; Step 4: Feature extraction and representation learning; Step 5: Intelligent Recommendation Engine.

2. The method for intelligent recommendation status reporting based on BERT algorithm in a domestically developed information technology environment as described in claim 1, characterized in that, Step 2 includes: Step 21: Collect user behavior data, status report data, and related text data from multiple data sources in the domestic IT innovation environment; Step 22: Clean, denoise, and standardize the collected data to improve data quality; Step 23: Perform natural language processing operations such as word segmentation and part-of-speech tagging on the text data to construct a data format suitable for the BERT algorithm input.

3. The method for intelligent recommendation status reporting based on BERT algorithm in a domestically developed information technology environment as described in claim 2, characterized in that, The user behavior data in step 21 includes filling in information, browsing, rating, liking, and saving.

4. The method for intelligent recommendation status reporting based on BERT algorithm in a domestically developed information technology environment as described in claim 1, characterized in that, Step 3 includes: Step 31: Build a BERT model training platform in the domestic IT innovation environment, and train the BERT model using the preprocessed dataset; Step 32: Employ gradient descent and attention mechanisms to improve the performance and accuracy of the BERT model in the context of information technology innovation; Step 33: Evaluate and validate the trained BERT model to ensure its stability and reliability in the information technology innovation environment.

5. The method for intelligent recommendation status reporting based on BERT algorithm in a domestically developed information technology environment as described in claim 1, characterized in that, Step 4 includes: Step 41: Use the trained BERT model to extract features from the status report text and convert the text into a low-dimensional vector representation; Step 42: Learn the feature representation of the status report, mine the key information and semantic features, and provide a basis for subsequent recommendations.

6. The method for intelligent recommendation status reporting based on BERT algorithm in a domestically developed information technology environment as described in claim 1, characterized in that, Step 5 includes: Step 51: Define a weight table for user behavior in the status reporting system, and configure the weights for user actions such as filling in information, browsing, rating, liking, and collecting.

7. The method for intelligent recommendation status reporting based on BERT algorithm in a domestically developed information technology environment as described in claim 6, characterized in that, Step 5 includes: Step 52: Combining the user behavior weight table and user behavior information, select N sets of status reports with the highest interaction degree from all the status report data that the user has interacted with: SET1: {CR1,…,CR N This represents the user's interest value.

8. The method for intelligent recommendation status reporting based on BERT algorithm in a domestically developed information technology environment as described in claim 7, characterized in that, Step 5 includes: Step 53: Based on the feature representations of the N status report data SET1 with the highest interaction weight scores selected in the previous step, apply the semantic matching method of the BERT algorithm to match the status report scores of the top M similarity scores to map MAP1:{{CR1,0.9},…,{CR M }}, and remove the status report data that has been interacted with by the user, to obtain a new ordered set of status report data MAP1':{{CR1,Score1},…,{CR M’， Score M’ }}, where Score = Similarity * Interaction Weight Score.

9. The method for intelligent recommendation status reporting based on BERT algorithm in a domestically developed information technology environment as described in claim 8, characterized in that, Step 5 includes: Step 54: Based on the user's department information, their job information in various systems within the company, and the time decay factor, adjust the sorting of status report scores in MAP1' to generate a personalized intelligent recommendation set of status report data for the user.

10. The method for intelligent recommendation status reporting based on BERT algorithm in a domestically developed information technology environment as described in claim 9, characterized in that, Step 5 includes: Step 55: Adjust and optimize the recommendation results based on users' real-time feedback and behavioral data to improve the accuracy and satisfaction of the recommendations.

Images