Machine Learning (algorithm)
Geographic Semantic Language Understanding and Spatial-Temporal Context-Aware Ranking for Location-Based Information Retrieval and Recommendation
Amap is a typical LBS scenario containing search, recommendation, voice assistant service, etc. Via our services, users can locate and navigate to POIs and take advantage of other helpful contents of POIs, such as opening time, comments and images. Furthermore, we offer several novel services such as hotel ordering, fuel charging and group purchasing.
Compared with the information services for other domains, geographic semantic, geographic locations and geographic knowledge base play a crucial role to query understanding and item ranking. Besides, taking account of the usage of map services are often related to trip or outgoing, personalized spatial-temporal prediction and ranking for recommendation are essential to optimize the accurate matching between information and users.
Our domain offers huge amounts of static and dynamic data, based on which the application of deep learning will benefit remarkably to the user experience and the development of our business.
- An embedding representation fusing geographic and semantic information in the field of geographic text processing, to improve the ability of geographic NLP understanding.
- A unified model of spatial location and text information to improve the geographic semantic relevance of query and POI in location retrieval.
- A spatial-temporal sequence model of user movement, to predict the personalized needs of users, and improve the recall and ranking of search/recommendation.
Related Research Topics
- Deep neural network architecture based semantic matching in information retrieval system.
- Exploiting POI-specific geographical influence for Point-of-Interest recommendation
- Content-aware hierarchical Point-of-Interest embedding model for location recommendation
- To improve accuracy and efficiency of multi-sentence scoring task by adopting poly-encoders
- To quantify the query-POI relevance by incorporating semantic similarity and geographical correlation.
- A dual-tower model to integrate dynamic situational context and users’ sequential behaviors in query-POI matching system
- Identifying linguistic areas for geolocation
- Spatial language representation with multi-level geocoding
Suggested Collaboration Method
AIR (Alibaba Innovative Research), one-year collaboration project.