Browse end-to-end machine learning pipelines built with PAI Designer, covering data preparation, feature engineering, model training, and deployment. Each use case links to a step-by-step guide you can follow directly.
Intelligent recommendation solutions
| Case | What you build |
|---|---|
| Recommendations based on object features | A product recommendation model driven by item attribute features. |
| Use FM-Embedding for recommendation recall | A recall pipeline that uses Factorization Machine (FM) and Embedding algorithms to generate feature vectors for users and items. |
| Create an FM recommendation model based on the Alink framework | A recommendation model built from the preset FM algorithm template in Designer. |
| Use collaborative filtering to recommend products | A product recommendation pipeline using the collaborative filtering algorithm. |
| Use the bipartite GraphSAGE algorithm for recommendation recall | User and item vectors for recommendation recall, generated with the bipartite GraphSAGE algorithm. |
| Use the ALS algorithm to predict music ratings (Old version) | Predicted user ratings for music using the Alternating Least Squares (ALS) matrix factorization algorithm. |
Intelligent risk control solutions
| Case | What you build |
|---|---|
| Implement public opinion risk control based on reviews from a food delivery platform | A risk management model that detects public sentiment from food delivery platform reviews. |
| Use graph algorithms to manage financial risks | A financial risk management pipeline built on graph algorithms. |
| Scorecard credit scoring | A credit scoring solution using the financial components provided by PAI and a scorecard model. |
| Abnormal metric monitoring | An anomaly detection model for monitoring system metrics. |
| Monitor user churn | A user churn prediction model built with PAI's user feature algorithms. |
Customized recommendation algorithms
| Case | What you build |
|---|---|
| Feature engineering | The feature engineering stage of a customized recommendation pipeline. |
| DSSM vector recall | A recall stage using the Deep Structured Semantic Model (DSSM) for candidate retrieval in a recommendation solution. |
| Sorting | The ranking and sorting stage of a recommendation pipeline. |
| U2I2I recall based on etrec | A U2I2I recall stage using etrec in a recommendation solution. |
Other general cases
| Case | What you build |
|---|---|
| Implement consistent offline and online CTR prediction | A unified click-through rate (CTR) prediction model trained on the Avazu dataset and deployed to EAS using the complete offline-tested Normalization Prediction->One-hot Encoding Prediction->Vector Assembler->FM Prediction workflow. |
| Predict heart disease | A heart disease prediction model using data mining algorithms. |
| Classify news based on text analysis algorithms | A text classification model for news articles, built with PAI's text analysis components. |
| Predict agricultural loan issuance based on regression algorithms | A loan issuance prediction model using linear regression on historical agricultural loan data. |
| Discretize continuous features using the Binning component | Discretized continuous features using the PAI Designer Binning component. |
| Population census statistics case (Old version) | A statistical model that analyzes how education level affects income, using population census data with attributes such as age, job type, and education level. |
| Predict student exam scores | A final exam score prediction model using logistic regression on middle school students' family background and school behavior data, identifying key factors that affect academic performance. |
| Automatically classify similar tags | An automatic classification system for product tags using PAI's text analytics component. |
| Predict hazy weather | A hazy weather prediction model trained on one year of real weather data from Beijing, identifying the pollutants with the greatest impact on PM2.5 levels. |
| Predict the output power of a power plant | A power plant output prediction model built from a preset workflow template in Designer. |
| Identify electricity theft | An electricity theft and leakage detection model built from a preset Designer workflow template, automating user checks to reduce manual inspection workload and help ensure normal and safe electricity use. |
| Offline scheduling | An offline scheduling pipeline using PAI's data mining components, demonstrated with an ad CTR prediction scenario. |
| Use TensorFlow to classify images | An image recognition prediction model built with the TensorFlow deep learning framework. |