Title: Recommending documents using Feature Guided Automated Collaborative Filtering
1Recommending documents using Feature Guided
Automated Collaborative Filtering
Gabriela Polcicová, Pavol Návrat
- Department of Computer Science and Engineering,
Faculty of Electrical Engineering and Information
Technology, Slovak University of Technology - polcicova_at_dcs.elf.stuba.sk
2Overiew
- Information Filtering
- Recommender system
- Limitations and open problems
- Conclusions
3Information Filtering
- Information Filtering
- delivery of relevant information to the people
who need it - Content-based Filtering
- recommending documents based on content and
properties of document - Automated Collaborative Filtering
- making recommendations using opinions of users
- Featute Guided Automated Collaborative Filtering
- making recommendations using opinions of users
accordig to categories
4Example of Information Filtering
Feature Guided Automated Collaborative Filtering
Collaborative Filtering
5Example of Information Filtering
Feature Guided Automated Collaborative Filtering
Collaborative Filtering
6Example of Information Filtering
Feature Guided Automated Collaborative Filtering
Collaborative Filtering
7Recommender System
- collects users opinions
- finds like-minded users
- makes recommendation
8Communication Agent
WWW
HTML-documents
Ratings
CA
URL-category
List of recom- mendations
Request for profile registration
User
List of recommen- dations
URL- -category
List of ratings
Request for profile registration
Re- commen- dations
Profile
Server
Yahoo!
URL-category
9Communication Agent
WWW
HTML-documents
Ratings
CA
URL-category
List of recom- mendations
Request for profile registration
User
List of recommen- dations
URL- -category
List of ratings
Request for profile registration
Re- commen- dations
Profile
Yahoo!
Server
URL-category
by the system
Category is determined
by the user
Rating URL, value, category
Recommendation URL, predicted value, category
10Communication Agent
WWW
HTML-documents
Ratings
CA
URL-category
List of recom- mendations
Request for profile registration
User
List of recommen- dations
URL- -category
List of ratings
Request for profile registration
Re- commen- dations
Profile
Yahoo!
Server
URL-category
List of predictions and like-minded users
List of ratings and found like-minded users
List of recommen- dations
Profiles of like-minded users
List of profiles
RA
List of profiles
Profiles of other users
11Limitations
- Collaborative filtering
- critical amount of rated documents is required
- recommends documents rated by at least one user
- Documents categorization
- categorization requires users interaction
- user can not change assigned category
Open Problem
- Using past predictions in computing correlations
- how much can this help?
12Conclusions
- Preliminary results
- recommendations are made by users agents
independently - in addition to ratings, previously computed
predictions are used - Future work
- using predictions from content-based filtering
13This method can be used
- By a community of users that
- communicate on Internet
- use electronic web documents
- rate documents
14Degree of Similarity of Two Profiles for Each
Category
15Prediction of the Rating