SIS: A system for Personal Information Retrieval and Re-Use

Goal of SIS
Finding and reusing previously used information.
Make it easy for people to find information they have already seen before.
Increase speed of search
Query refinement
New ranking ideas which are more personal to user

Key aspects of SIS
Unified index to information from all kinds of sources on a computer
Since information has been seen before, we have rich contextual cues obtained can be used in the searching and presenting information

System architecture

User Interface
Top View - filters for refining attributes in each coloumn

User Interface (ctd)
Side view - simplified filters. Eg. Outlook express

Evaluation
Log data - Detailed information on the nature of user queries, interactions with the UI and properties of items retrieved.
Questionnaire data - asking questions on how people organized their data before and after using SIS and about their experiences.

Observations
25% of queries involved people’s names, indicating people are a powerful memory cue for personal content.
Most query types were People/Places/Things, Computers/Internet & Health/Science.
Filters used were file types and date.

Observations using log data
Graph shows that recent items are accessed more frequently than others.

Observations (ctd)
We also see that frequency of access of items decreases since the time they are created. Email has the steepest graph since it has a shorter effective life than other documents.

User Interface observations
Top view was preferred to side view
Most users sorted the information by date and rank. This shows that many searches were made over personal content. Date is more useful over other attributes for sorting personal items.

Observations of questionnaire data
The graph shows that the ease of finding information increased. Also there was a decline in non-SIS searches after people were introduced to it.

Review of paper
No information on the implementation of their software
More focus on the experiments and observations
Would like more information on the unified index structure.
Nothing novel about the user interface
http://news.com.com/2009-1032-1020641.html MSNbot

Cognitive Expansion
What information is relevant for expansion - sources of information are user’s long term preferences, intention, situation and knowledge of specific domain
How to use the information

Implementation
Preference Manager - administers all explicit query preferences for each user and stores them in a suitable repository for future use
P-Inference Engine decides what preferences to choose for expansion depending on user profile and other information sources mentioned earlier.

Architecture