A Location Predictor based on Dependencies Between Multiple Lifelog Data Masaaki Nishino Takashi Yagi Yukihiro Nakamura Shinyo Muto NTT Cyber Solutions Laboratories, NTT Corporation Masanobu Abe Graduate School of Natural Science and Technology, Okayama University ACM LBSN 10, November 2, 2010, San Jose, CA, USA 1
Location Prediction By predicting future locations, we can provide useful information to a user. To-do tasks related to the place Time limited Sales Time limited sale info at the supermarkets near the predicted locations Weather info of future locations To-Do Tasks related to future locations 2
Related Works Extract and use regularities in movement from location log Markov Model[Ashbrook, 2003] Dynamic Bayesian Network[Liao, 2007] Sequential Pattern Mining[Monreale, 2009] Office 0.6 1.0 1.0 Home accumulated location logs Station 0.4 prediction models 3
Problems Since previous research uses regularities, they cannot predict irregular movements. Regular go to school on weekdays go to the gym every Monday etc... Irregular irregular meeting business trips etc... 4
Our proposal Integrate different kinds of lifelogs to predict both regular and irregular movements Regular go to school on weekdays go to the gym every Monday. etc... Irregular irregular meeting business trips can predict with location log. can predict with integrating different kinds of logs. 5
Calendar data We use calendar data as a source for making predictions of irregular movements People enter info about irregular events into it Widely used. 6
Dynamic Bayesian Networks model for integrating different data. DBN model can make reasonable predictions when prediction with only calendar/gps data is difficult Stay Duration 10 min 0 min 15 min GPS Data Location Office Office Bldg A Calendar Lunch Bob Time of Day Evening Evening Evening Calendar Data Integrated DBN Model 7
Preprocesses Stayed places (37.53,-122.08) (37.50,-122.00) (Office) (Station A) (Office) (Station A) GPS Data Clustering Stayed Places Meeting with Bob See the Dentist Meeting, Bob Dentist Meeting Bob Dentist Calendar Data Extracting Keywords Keywords 8
Concepts of DBN Model Place-place relationship Markov Model Home Office Building A Location Simple model for predicting locations. Can predict regular movements. 9
Concepts of DBN Model Place-Keyword relationship Infer a user s own relationship between place and keyword from co-occurrences. Building A Meeting Bob Location Calendar Can predict irregular movements 10
DBN model (basic) Integrate place-place relationship and place-keyword relationship Location Home Office Building A Calendar Lunch Bob Both regular and irregular movements can be predicted. 11
DBN model (actual model) Make some extensions to basic model. - add node that represents time of day, stay duration Stay Duration 10 min 0 min 15 min Location Office Office Bldg A Calendar Lunch Bob Time of Day Evening Evening Evening 12
Learning Estimating the parameters of the probability distributions of DBN from data. Using maximum a posteriori (MAP) estimation. 13
Inference Use the Viterbi algorithm to infer a state sequence that maximizes probability. 10:00 10:10 10:20 10:30 10:40 10:50 Stay Duration Location Initial State Calendar Time of Day 14
Experimental Settings Whether prediction accuracy is improved or not by using calendar data. Baseline: DBN model without Calendar data Dataset GPS and calendar data of two subject (about 50 days) 15
Evaluation metrics Evaluate the accuracy of prediction by changing the time difference between the subject of prediction and the time to start prediction Building A Time to Start Prediction Home Station B Station A Subject of the prediction Time (ago) 10:00 (120) 11:00 (60) 12:00 (0) 16
Prediction Results Regular Movements Subject A 0.7 0.8 Subject B 0.6 0.75 0.7 0.5 0.65 0.4 0.6 0.55 0.3 0.5 200 150 100 50 0 200 150 100 50 0 GPS GPS+Calendar Blue line Red line There are no much differences. 17
Prediction Results 0.6 Irregular Movements Subject A Subject B 0.35 0.5 0.3 0.4 0.25 0.3 0.2 0.2 0.15 0.1 200 150 100 50 GPS GPS+Calendar 0 0.1 200 150 Blue line Red line 100 50 0 The accuracy was improved for irregular movements 18
Example of Predictions The prediction failed since the wrong place is estimated from the calendar entry 19
Example of Predictions The result was modified because the time needed for movement was considered 20
Conclusion We show a DBN model for making prediction for both regular and irregular movements by using GPS and calendar data. The accuracy of predictions for irregular movement was improved. Wrong prediction due to wrong schedule can be modified by using GPS data. Future works Use other kinds of logs. 21
Thank you. 22