user behavior analysis of location aware search engine third international conference of mdm, 2002...
Post on 03-Jan-2016
216 Views
Preview:
TRANSCRIPT
User Behavior Analysis of Location Aware Search Engine
Third international Conference of MDM, 2002
Takahiko Shintani, Iko Pramudiono
NTT Information Sharing Platform Lab.
Summarized by 공기현
2008.07.17
Copyright 2006 by CEBT
Introduction
Access log of a web site records every user requests
From the Access log, we can know Which pages were visited by the user
What kind of Requests submitted
Where the user come from
This paper focus on mining the behavior of user with re-gard to his location from user access log
We use association rule mining and sequential pattern mining for user log analysis Association rule mining
Sequential pattern Mining
IDS Lab. Seminar - 2Center for E-Business Technology
Copyright 2006 by CEBT
Mobile Info Search
MIS is a research project conducted by NTT lab
“Personalized digital guide portal”’ site services for mo-bile user
Provides location aware information from the internet by collecting, structuring, organizing and filtering
Between Users and information sources, MIS mediates database type resources such as online maps, internet “yellow-pages”
Authors collect user logs from MIS site
IDS Lab. Seminar - 3Center for E-Business Technology
Copyright 2006 by CEBT
MIS Functionalities
Location Oriented Meta Search provides a mediation ser-vice for database-type resources
Location Oriented Robot-based Search, “kokono”, pro-vides the spatial search that documents close to a loca-tion
IDS Lab. Seminar - 4Center for E-Business Technology
Copyright 2006 by CEBT
User Location Acquisition
The user location represents the geographical position, or the area of the information in the form of address strings (latitude, landmarks,…)
The user location is automatically obtained by Mobile Device such as GPS, PDA, Notebook
In this paper, we use PHS system and its Logs PHS use many small base stations
The base stations are placed in almost every stations, buildings, and street.
– User Location accuracy is better than Cell phone.
IDS Lab. Seminar - 5Center for E-Business Technology
Copyright 2006 by CEBT
kokono Search
How to collect Local Information? Robot gathers web documents from the Internet
Parser parses the obtained documents to look up the location in-formation
(address) and spatial information(longitude-latitude)
Store web documents with local information to repository
How to structure the Local Information? Divide document into morphemes by the parser
Compare noun phrase to the address dictionary and regard it as an address if it satisfies the following condition
– Any address strings without upper address
– Cities with address suffix (ex. Yokohama Shi)
– Towns or block numbers with the city name
– Block
IDS Lab. Seminar - 7Center for E-Business Technology
Copyright 2006 by CEBT
Mining MIS Access Log
Site statistics
Preprocessing Remove directly accessed log, Image retrieval and Back action for
valid analysis
IDS Lab. Seminar - 9Center for E-Business Technology
Copyright 2006 by CEBT
Access Log Format
Each search log consists Web CGI parameters Location information (Address, station, zip, …)
Location acquisition method ( from)
Resource type (submit)
Name of resource to search form ( shop, map, rail, station..)
Condition of search
Access Hour, Access Date
IDS Lab. Seminar - 10Center for E-Business Technology
Copyright 2006 by CEBT
Transformation to Transaction table
Representation of access log in relational Database
IDS Lab. Seminar - 11Center for E-Business Technology
Copyright 2006 by CEBT
Experiment Result – Association Rule Min-ing
Results of User log mining regarding Search Condition
IDS Lab. Seminar - 12Center for E-Business Technology
Copyright 2006 by CEBT
Experiment Result – Association Rule Min-ing
Results of User log mining regarding time, location ac-quisition method
IDS Lab. Seminar - 13Center for E-Business Technology
Copyright 2006 by CEBT
Experiment Result – Sequential Rule Mining
IDS Lab. Seminar - 14Center for E-Business Technology
Copyright 2006 by CEBT
Conclusion
We reported the result of mining web access log of Mo-bile Info Search
We use two techniques, the association rule mining and sequential pattern mining
Using those two techniques, we can figure out how the behavior of MIS user and services they use are affected by their location
Unfortunately, there are many case when the user is overwhelmed by so many result Clustering the search results on their contents is required
IDS Lab. Seminar - 15Center for E-Business Technology
top related