1. Publish/Subscribe for textual data stream (with or without geo-locations)  

Massive amount of text data (with or without geo-locations, e.g., geo-tagged tweets) are being generated at an unprecedented scale. Users may want to be notified of interesting geo-textual objects continuously. We develop indexing and query processing techniques to handle a large number of subscription queries over geo-textual data stream or textual data stream. 

Publications:

-  Diversity-aware top-k publish/subscribe on text stream (newSIGMOD 15)
-  Temporal spatial-keyword top-k publish/subscribe on geo-textual data stream (
newICDE15 and demo in VLDB14)
-  Boolean spatial-keyword publish/subscribe on geo-textual data stream (SIGMOD13  and demo in VLDB14)


2. Spatial-keyword querying 

There exits a large volume of geo-textual data (e.g., geo-tagged tweets and POIs in Foursquare) featured with both textual and geospatial content. To efficiently process spatial-keyword queries on such data, we develop new indexing techniques and query processing algorithms. 

Publications:

-  Answering the m-closest keywords query (newSIGMOD 15)
-  Search regions of interest for user exploration  (VLDB14)
-  Distributed spatial keyword querying on road networks (EDBT14)
-  An evaluation of 12 geo-spatial indexes (VLDB13). Code available here
-  An overview paper on spatial-keyword querying (invited paper in ER)
-  Route planning: answering queries like “a most popular route such that it passes by shopping malls, restaurant, and pub, and the travel time is within 4 hours.” (PVLDB12)
-  Efficient processing of several types of spatial keyword queries (VLDB09, PVLDB10, SIGMOD11a).  Code for our SIGMOD11 paper is available here.  An extension of our SIGMOD 11 paper is published in TODS (new)
-  Efficient algorithms and cost models for reverse spatial-keyword k-nearest neighbor search (SIGMOD11b, TODS14)
-  Efficient spatial keyword search in trajectory databases  (unpublished paper)


3. Mining users' spatial-temporal behaviors and context-aware POI recommendation

Micro-blogging services and location-based social networks, such as Twitter, Weibo, and Foursquare, enable users to post short messages with timestamps and geographical annotations. The rich spatial-temporal-semantic information of individuals embedded in these geo-annotated short messages provides exciting opportunity to develop many context-aware applications in ubiquitous computing environments. We develop models to discover individual users’ spatial-temporal behaviors, which refer to who visits which place at what time for what activity. That is, we consider user (Who), spatial (Where), temporal (When), and activity (What) aspects. The models have a variety of applications in contextual-aware recommendation and search . We also develop new methods for POI recommendation. 

Publications:

-   Our recent work propose a new POI recommentdation approach, which performs better than previous approaches in your experiments (newSIGIR 2015)

-   SAR: A sentiment-aspect-region model for user preference analysis and POI/user recommendation. The model provides explanations for recommendation results.  (newICDE 2015)
-   A general graph model for recommendation in heterogeneous networks and its applications in event-based social networks (newICDE 2015)
-   Diversity-aware POI recommendation (newAAAI 2015)
-   Group Recommendation (KDD14)    
-   W4: Discovering spatio-temporal topics for individual users and its various appliations, e.g., requirement-aware POI recommendation  (KDD13, newTOIS15).   Datasets available here
-   Time-aware POI recommendation (SIGIR13, CIKM14).  Datasets available here
-   Mining significant semantic locations from user generated GPS data for recommendation (PVLDB10)


4. Finding influential users in social networks

We develop efficient algorithms for discovering influential users in social networks. In particular, we consider the impact of users' attributes,  time factor, and novelty decay (Repeated exposures of an individual to an idea may have diminishing influence on the individual) for finding influential users. 

Publications:

-  DynaDiffuse: A dynamic diffusion model for continuous time constrained influence maximization (newAAAI 15)
-  Finding influential event organizers in event based social networks (SIGMOD14)
-  Influence maximization with novelty decay (AAAI14)
-  Time constrained influence maximization in social networks ( ICDM12 , TKDE . Source code), 
-  Computing top-k influential nodes (KDD10, AAAI 11)  


5. Mining reviews, social media, and forums

We develop techniques for review mining and sentiment analysis. We also develop techniques for mining social media, including Micro-blogs (e.g., Twitter), and Community Based Question Answering Sites (e.g., Yahoo! Q&A).

Publications:

-   Detecting user intents from tweets (newAAAI 15)
-   Coarse-to-fine review selection via supervised joint aspect and sentiment model (SIGIR14)
-   One seed to find them all: Mining opinion features via association (CIKM12)
-   Geolocation prediction for social images by exploring user profiles (JASIST14
-   On predicting popularity of newly emerging hashtags in Twitter (JASIST13)
-   Short text classification ( WWW12 poster, evaluation paper JASIST ) and hierarchy maintenance ( SIGIR12).  Annotated dataset  for our SIGIR12 paper is available here.
-   Using categorization information to improve question search in community based question answering services ( CIKM09, WWW2010, TOIS12). Annotated dataset is available here
-   Extracting Question-Answer pairs from forums to build the QA database (SIGIR08, ACL08)
-   Routing questions to expert users ( ICDE09)


Acknowledgement: Some of these projects are supported by grants awarded by Ministry of Education, Singapore, and a grant awarded by Microsoft.