Elsweiler, D. and Ruthven, I. and Jones, C. Dealing with fragmented recollection of context in information management. In: Context- Based Information Retrieval (CIR-05) Workshop in Fifth International and Interdisciplinary Conference on Modeling and Using Context (CONTEXT-05).
Dealing with Fragmented Recollection of Context in Information Management David Elsweiler, Ian Ruthven, Chris Jones University of Strathclyde Context-based Information Retrieval CIR-’05
Motivation I read about that somewhere but where did I read it? I know I have a document that would help with that but I cannot think of where to find it. I need a good picture of …. But it will take me ages to search through my collection I’ve been to a webpage that had information on that but I will have to search for it.
Memory and PIM Memory mediates interaction Re-finding Information Objects governed by recollection Internal Context External Context Previous work suggests that recollections are rarely complete Partial fragments of many types Tools we have burden memory The way they work isn’t aligned with what we remember
An Interface for re-finding Photos What do we remember about photographs? Study of recollection of personal photos 9 subjects, 12 random personal images Annotation Textual Descriptions Group Images Recollection After delay 1 week – free recall session Role of cues
Aims What features were annotated? Did users tend to annotate the same features? Did free recall performance vary? Did recollected descriptions match annotations? Did performance vary with cues? Outcomes Annotations varied greatly Features and detail Varied across population Varied across photos for individuals
Annotations (Features and Groups) Subject (primary object(s) or actor(s)) By name, relationship to the subject, subject opinions, associated facts Environment Brief general descriptions, weather, facts about environment Events / Activity Where / what was happening Temporal Information Date (mostly year), often linked to event / activity
Recollection findings Free recall performance was good 7.78 images from 12, st.dev = 1.47 Probably decrease with time and larger collections Recalled descriptions varied hugely Some participants described recollected images in greater detail than their annotations Users often recalled annotation process
Main Outcomes Annotation can elaborate encoding But unlikely to be able to recreate annotations Underlined Heterogeneous nature of memory We all remember different details Depending on many factors We need tools to support this
Evaluating PhotoMemory (pilot) Compare PhotoMemory (system 1) with: A reduced version of software (System 2) Hierarchical folder structure (System 3) Aims Find evidence for / against our theories Learn more about recollections and how they are used Learn about ways to improve the implementation of the theories
Evaluation Methodology 6 participants Own hierarchical folder structures of images Number of images (mean 207) Given 3 weeks with systems Observed & Timed subjects on 3 task types Photographs from particular event Individual photograph from the collection Photographs that spanned different events Surveyed Pre and Post – experiment Pre and Post task
System Performance TaskSystem1System 2System3 Task 1 Time to complete task (secs) No. images retrieved Selected group accurately reflected recollections Task 2 Time to complete task (secs) No. images retrieved1.00 Selected group accurately reflected recollections Task 3 Time to complete task (secs) No. images retrieved Selected group accurately reflected recollections Average for All Tasks Time to complete task (secs) No. images retrieved Selected group accurately reflected recollections
System Ratings System 1System 2System 3 Preferred System600 Easiest to system to use321 Fastest System600 Most effective when searching for one photograph600 Most effective when searching for multiple photographs600
Observed Behaviour Recollected Features Contextual Fragments Often Highly Visual – “rustic pink coloured railings with snow lining the top” Person Feelings & Emotions – “we were so cold and tired” Relationships Between Images – “one was earlier – it was warmer and sunnier, while another was later – you can see that we were getting cold”. Recollections appeared to influence search behaviour, but not all fragments were used
Observed Behaviour (cont) Bursts of Searching (on all 3 systems) Focus on Single Aspect of Image they Remembered Only when a burst failed did they use other fragments PhotoMemory occasionally allowed multiple fragments in single search burst Increasing Recollection
Observed Behaviour (cont) Dynamic Groups Apply filters to create groups of images Referred to such sets as having shared attributes Treated as if they were folders in hierarchy System 3 (hierarchy) Performance depended on both recollection of photograph(s) but also knowledge of space
Conclusions Memory has a major role in re-finding photos Users tend to recall partial contextual fragments Remember more context than they use when re-finding If you provide tools that enable them to use more context they will Seems to improve performance Seem to like it