Presentation on theme: "Software-based Assistant for Personal Information Management By: Nuno Magalhaes Ribeiro Hypermedia & Multimedia Group Department of Computer Science The."— Presentation transcript:
Software-based Assistant for Personal Information Management By: Nuno Magalhaes Ribeiro Hypermedia & Multimedia Group Department of Computer Science The University of York email@example.com Supervisor: Dr. Ian Benest Assessor: Dr. Patrick Olivier December 14, 1998
14/12/98Department of Computer Science The University of York 2 Motivation I enter my office for the first time today. My computer screen is blank as always. My computer senses me and asks me (through speech) if I want to read my new e-mail. I say Not right now, I have an important call to make first ! I dial the number and wait for the connection while thinking of everyone I must call to tomorrows important project meeting. Id like to have every thinking brain with me to help me decide what to do next, I think. In a moment I am on the phone with Paul, the project leader of my group, telling him the names of everyone I want to meet …Julia, Mike, Joan,… Suddenly, my computer screen displays a space ship moving around and the four letters that compose a foreign name…JOSE Oh..and Paul, please dont forget to also invite that new project member, Jose, you know the one that wrote that interesting report about spaceships fuels. O.K. Bye, bye. See you tomorrow. I handle the phone and briefly think I almost forgot to tell him about Jose, before leaving my office thinking about the ideas I read in that report and how to make the most of them…
14/12/98Department of Computer Science The University of York 3 Identifying the Problem [Maes, 97] mismatch between the complexity of our lives and our cognitive abilities: too many things to keep track of information overload learn and remember more information The amount of (personal) information is increasing at a fast pace: difficult to remember information (what we know but cant recall) difficult to find existing information (what we need but cant find) time consuming for us to manage every piece of information we gather
14/12/98Department of Computer Science The University of York 4 An approach to tackle the problem [Bush, 45] A record…must be continuously extended, it must be stored, and above all it must be consulted. [Lamming et al., 94] a memory prostheses should become a companion in everyday life, automatically capturing data that may be of use later [Bell, 97] …one can imagine…a guardian angel that can capture and retrieve everything we hear, read and see. [Maes, 97] prosthetics for the mind (memory augmentation devices) in order to overcome our cognitive limitations: poor memory for details only deal with one thing at a time slow to process large amounts of information
14/12/98Department of Computer Science The University of York 5 Notion of a Shadow Assistant A shadow assistant stays in the background and observes what we do without interfering, learns about us and our tasks and augments our memory. 1] Observes the user and gathers information: monitors the information the user is collecting captures relevant data about the context of users tasks 2] Uses this information to assist: offers to automate mechanistic tasks helps recall important details about information needed for a task suggests relevant information to a task [Bush, 45], [ Lamming et al, 94], [Maes, 97], [Bell, 97]
14/12/98Department of Computer Science The University of York 6 Why is context important in PIM? [Lansdale, 88] poor memory for details. Therefore, systems which rely on the user to remember details such as filenames are bound to produce low levels of recall. [Lansdale, 88] Recall will be best when the cues present at the time of learning and at the time of recall are most alike. [Tulving, 83] (Encoding Specificity Principle) physical location of an event, who was there, what was happening at the same time and what happened imediately before and after. [Tulving, 83] presenting partial context information about an episode helps people remember more about it. [Tulving, 83] (reconstructing the context). associate our current environment with past experiences that might be related, in order to suggest existing relevant information to our current task. [Rhodes, 97] (context matching).
14/12/98Department of Computer Science The University of York 7 What is a Personal Assistant? [Hoschka, 96]...some functions of good human assistance [Lennon and Vermeer, 95] …It thus supports the user at all levels of activity…making predictions from repetitive tasks, it saves us both time and frustration…it will be our augmented eyes and ears, an alter ego we create for ourselves Assistant Required Properties Domain Competence Competence Assessment Learning and adapting behaviour Processing imprecise instructions Explaining abilities Cooperation support
14/12/98Department of Computer Science The University of York 8 What is an Automated Personal Assistant? [FIPA, 97] Overview of the Personal Assistant Domain Software Agent: acts semi-autonomously on behalf of a user (wo human guidance) models the interests of the user, provides services to the user or other people/personal assistants, accomplishes routine support tasks (real job), is unobtrusive but ready when needed.
14/12/98Department of Computer Science The University of York 9 What is an Automated Personal Assistant? Interface Agents learn from, act on behalf of and collaborate with the user [Maes, 94] and Information Agents look, find and compose information [Nwana, 96] An Automated Personal Assistant is a software agent that both exhibits Interface Agents properties and performs Information Agents tasks.
14/12/98Department of Computer Science The University of York 10 Analysis of existing approaches to capture users activities context Issues: 1] What contextual information about users tasks is gathered? 2] How is this contextual information used? Survey brings together a set of papers which common purpose is: to collect context elements, to provide ways to explore these elements to help retrieving information (once known but know forgotten).
14/12/98Department of Computer Science The University of York 11 What context information about users tasks is gathered? Events are used to describe context of tasks Time-stamp on documents and events Operations performed on documents /files (edit, save, create, print, exchange) Physical events (people met, location, weather report) Communication events (e-mail received / sent, phone number called, news) Manual annotations (notes, diary entries for meetings, schedules, visits) Textual context (word vector for text documents, subject of e-mail messages) Memoirs [Lansdale and Edmonds, 92] Forget-Me-Not [Lamming and Flynn, 94], [Lamming et al., 94] Hive [Bovey, 96] Lifestreams [Freeman, 96] Remembrance Agent [Rhodes, 97], [Crabtree and Rhodes, 98]
14/12/98Department of Computer Science The University of York 12 Analysis Common characteristic: time temporal relations between events helps recall. although precise data about an episode is not usually remembered, temporal relations between episodes are remembered very well [Lamming et al., 94]. The more information a system collects about context of users activities, the more help it can be to recall details. Forget-Me-Not, Hive,Remembrance Agent a system that is sensitive to context (location and people) can provide reminders / suggestions appropriate to current activities. Indexing documents by context improves recall (document retrieval). context as a retrieval key (only events / groups of events, not tasks)
14/12/98Department of Computer Science The University of York 13 How is contextual information used? 1] Retrieve by context reconstruction (you tell what you remember) User searches for information by specifying remembered context details, System matches users details with existing stored context elements, System presents users with a subset of events that matches details, Users browse existing events and select those relevant. 2] Suggest by context matching (you automatically receive suggestions) Remembrance Agent, pro-actively matches current user context with stored contexts about past experiences, makes automatic suggestions (only works for text processing). Useful utilisation of context elements, particularly in the case of pro- active suggestions if we can extend the idea to activities other than text processing.
14/12/98Department of Computer Science The University of York 14 Critique What we didWhy we did it What resulted How we did it What did we want When we did it What problems did we solve What caused it Who was thereWhere we wereWhat was happening It is possible today, but not extremely useful Index by Events It is much more difficult, but also much more useful because it would allow the shadow assistant to know what we were doing Index by Tasks High Level Low Level Is it possible to infer High Level events from Low Level events?
14/12/98Department of Computer Science The University of York 15 Open Issues How to infer users tasks from low-level events? What is relevant information about each users context? Do we want to be constantly observed? (Privacy) What if our personal information falls in the wrong hands? What happens if we can remember everything all the time? [Bos, 95] Where can observations take place? Do we want a humanised interface?