Friday, February 23, 2007

PAWS Meeting 2007-02-23

1. Personalized Course Navigation Based on Grey Relational Analysis

Paper presentation by Sharon

This paper introduced a study on a personalized course navigation based on a measure called WGRA (Weighted Grey Relational Analysis). It aims to let users browse open corpus educational Web resources and recommend them according to user interests. The task is done by the WGRA measure and implicit feedback information. It can provide easily maintained user models, low complexity for the computation required, and ease to add knowledge.

Even though this study seems rather simple, but it has a strong relevance to some of the projects done within our group such as KnowledgeSea, so it was recommended to the persons to check the paper.

2. Unsupervised and Supervised Machine Learning in User Modeling for Intelligent Learning Environment

Paper presentation by Tomek

The motivation here is to reduce the effort for constructing user models and improve transferability across applications. The proposed framework takes two learning steps, unsupervised and unsupervised in order to model students' learning during interaction with an education support systems (exploratory learning environment). The first step is done by clustering, where users' interaction behaviors are fed in. After the experts examined the clusterings, they label each of them according to the students' achievements and use them in the next step, online classification (k-means).
The ideas for applying the proposed approach to relevant projects were discussed and this paper was also recommended.

Friday, February 16, 2007

PAWS Meeting 2007-02-16

1. First Paper
Rosta presented a paper "Accurately Interpreting Clickthrough Data as Implicit Feedback". The authors investigate users' behavior given a Web search task. The data collected involved eye-tracking data. Questions that the paper addresses are:

- Which links do users click (as opposed to only look at)?
- What does each click mean (does it mean?
- Do users scan top to bottom?
- Which links do they evaluate before clicking?

2. Second Paper
Danielle presented a peper "ResultMaps: Search Result Visualization for Hierarchincal Information Spaces". This visualization is based on a Treemap. It presents search results in the context of full information space. A "global" ResultMap is showed next to regular textual list of search results. Each textual result can be expanded to see its contribution to the "global" ResultMap (a "local" ResultMap is presented). Each ResultMap presents resources of different type (e.g. article, lecture, video) with different color. The sizes of map cells does not carry any information. User performance evaluation and enhencements of the representation is on the authors future research agenda.

3. Housekeeping
We discussed reviving our Wiki. The proposed structure looks as follows:

Relevant Groups
PhD Theses
Opprtunities (Companies, Jobs etc.)
Community Services (Tools etc.)

Jae-wook will create the structure. All PAWS members are responsible for adding content relevant to their area of focus. We may assign people responsible to different categories later.

Friday, February 09, 2007

PAWS Meeting 2007-02-09

1st talk: Michael Yudelson
Topic: Recommenders for Information Seeking Tasks: Lessons Learned.
Goal of the recommender is to meet user specific needs with respect to:Correctness ,Saliency,Trust,Expectations and Usefulness
General Advice :
o1.Support multiple information seeking tasks
2.User-centered design nShift focus from system and algorithm to potentially repeated interactions of a user with a system
3.Recommend nNot what is “relevant”, But what is “relevant for info seeking task X”
Choice of the recommender algorithm: Saliency (the emotional reaction a user has to a recommendation) , Spread (the diversity of items) , Adaptability (how a recommender changes as a user changes) , Risk (recommending items based on confidence)

experiment results found out:
Bayes - generated similar recommendations for all users and PLSI - generated random, “illogical” recommendation

very good papers for everyone.

2nd talk: Jae-wook Ahn
Topic: A Novel Visualization Model for Web Search Results.
-what's missing with current approaches? semantic relations/views, degree of relevance in terms of subjects of interest
-speed and subject of interests are new in this research.
-the system list of keywords for user to manipulate the their interests. weights are adjustable.
-WebSearchViz system, will release the opensource soon. so far, not so much detail demostration on their web.

how does the rotation really work?
profile the keywords has been done long time ago, it's been moved to profile the concepts

universal portal for centralized login discussion:
benefit from each system's user model.
Chirayu talked about the RBAC backend role based server as an option.
how to use existing technology to integrate all these?