- 陆奇(微软执行VP)
- information, service, interaction
- linked web graph, real social graph, precise spatial graph, realtime temporal data(sns)
- natural UI: in-context, proper
- challenges: nlp, intent, user model, knowledge representation and acquirement, QA, task model for domains, machine learning and feedback
- CIKM2014 Cup
- word2vec
- keyword、entity
- query expansion
- session query
- Gerhard Weikum(数据库和知识库大牛, 关于knowledge graph)
- sources: dbpedia、yago、freebase
- entity:relation、category、linkage、popularity、context
- inference:desc、relation、popularity=>densest graph
- expert data vs social media
- long tail entity in social media
- application: context-aware suggestion
- application: privacy risk advisor
- Jeff Dean(Google senior fellow)
-
- general model for all tasks
- 并行化是关键思想, 通过层内只做局部连接
- dense data靠增加层数,sparse data靠embeddind
- application: step by step network for translation
- application: 生成模型做图像理解
CIKM2014参会印象
最新推荐文章于 2021-11-17 19:00:00 发布