基于時(shí)空大數(shù)據(jù)的推薦系統(tǒng)研究
時(shí)間：2022年09月22日（周四）上午10:00
地點(diǎn)：計(jì)算所948會(huì)議室
報(bào)告人：韓鵬
報(bào)告人簡(jiǎn)介：韓鵬，博士。丹麥奧爾堡大學(xué)計(jì)算機(jī)學(xué)院助理教授。電子科技大學(xué)協(xié)議副研究員。主要從事數(shù)據(jù)挖掘和人工智能方向研究，涉及領(lǐng)域有時(shí)空大數(shù)據(jù)、藥物發(fā)現(xiàn)、自然語言處理和計(jì)算機(jī)視覺。曾獲ImageNet視頻中的目標(biāo)檢測(cè)競(jìng)賽世界第二名，IJCAI-Alibaba商品推薦比賽世界第四名。已發(fā)表學(xué)術(shù)論文20余篇，其中CCF A類論文16篇，長(zhǎng)期擔(dān)任KDD、ICLR、AAAI、IJCAI、TKDE等會(huì)議和期刊審稿人。
Abstract：Point of interest (POI) recommendation has become an increasingly important sub-field of recommendation system and aims to find new places for users that they might be interested in. It can help users find interesting spots that will make them enjoy their vacations when they are in unfamiliar regions. And it can also increase the shopkeepers’ income by attracting more customers who would like to spend time and money at the store. Therefore, POI recommendation has become a hot research topic in recent years. However, there are many challenges in this problem and one of the most challenging one is the data sparsity problem. To tackle this problem, many methods incorporate the contextual information into the recommendation method with different assumptions. For example, some work assumes that the user will visit new POIs that are close to the POIs they visited before. And they construct an auxiliary label matrix by adding the weighted sum of neighboring POIs’ ratings to every POI. Some study assumes that users will have different preference patterns in different time slots, so they construct different models for different time intervals. Although the assumptions are various, the common property behind them is that similar users should visit similar POIs and similar POIs should be visited by similar users. Therefore, inn this seminar, how to utilize contextual information in the POI recommendation will be introduced with different frameworks by constructing similarity graphs.