Adversarial contrastive with leveraging negative knowledge for point of interest sequence learning

Area of Science:

• Data Mining
• Machine Learning
• Artificial Intelligence

Background:

• Learning user preference representation is crucial for mining point of interest (POI) data.
• Existing POI sequence learning methods often lack generalization and struggle with multiple downstream tasks due to end-to-end training.
• Current contrastive learning approaches in POI sequence learning overlook the valuable characteristics within negative samples.

Purpose of the Study:

• To propose a novel model, Adversarial Contrastive with Leveraging Negative Knowledge (ACLNK), to improve the generalization and performance of POI sequence learning.
• To enhance the ability of POI sequence learning methods to support multiple downstream tasks.
• To effectively utilize information from negative samples in contrastive learning for POI data.

Main Methods:

• Designed an adversarial generalizing representation module to capture long-term user preferences, incorporating social circle information.
• Developed a negative sample knowledge extraction attention mechanism to extract comprehensive short-term preferences from limited input sequences.
• Integrated learned short- and long-term preferences into a contrastive module to generate accurate user generalization representations.

Main Results:

• The proposed ACLNK model demonstrated significant improvements in effectiveness and generality across three check-in sequence datasets.
• ACLNK outperformed previous state-of-the-art models in two distinct downstream tasks.
• Extensive experiments validated the superior performance of ACLNK in POI sequence learning.

Conclusions:

• ACLNK effectively addresses the limitations of existing POI sequence learning methods by improving generalization and performance.
• The model's ability to leverage negative knowledge and incorporate social information enhances user preference representation.
• ACLNK offers a robust solution for diverse downstream tasks in POI data mining.