collaborative search techniques submitted by: shikha singla mit-872-2k11 m.tech(2 nd sem)...
TRANSCRIPT
Submitted By: Shikha Singla
MIT-872-2K11M.Tech(2nd Sem)
Information Technology
Personalized Search Systems
There is a large gap between how well search engine could perform if results were refer to the individual,& how well they currently perform by returning results to satisfy everyone. This gap is called Potential For Personalization.
Relevance Judgments
INTRODUCTIONCollaborative search systems are the systems which
supports group based searching, Like in knowledge work & education work.
Understanding the similar properties of people involved in group search sessions has the potential to significantly improve collaborative search systems.
To know about the similar properties of different people, we use potential for personalization.
Group Formation
Group Longevity Task based Groups(short term groups)
Group members are working together to accomplish this shared task, like- travel planning, shopping etc
Trait based Groups( long term groups)
Long-term groups are comprised of users who are related through shared traits, like-geography, occupation(job role or job team groups), demographics etc.
Group IdentificationExplicit Groups-
Group membership can be determined is by information provided directly from the members. Like-gender, age, geographic location, job-role etc.
Implicit Groups-
Group membership can be inferred from member activity ,based on their actions. Like- similar desktop indices, grouping users who issue similar search queries etc.
Groupization Group members’ data is used to rank an individual’s
search results by giving higher weights to pages that are relevant to more members of the group.
To perform groupization on a set of search results :- First calculate a personalization score for each search
result for each member of the group. Then for each search result, the groupization score is
computed as the sum of the personalization scores of each group member.
Then take a weighted combination of the groupization score and the search result’s original rank.
In this, we computed the normalized Discounted Cumulative Gain (DCG)(eg-for any query)
From this we can see that, the use of group data in addition to an individual’s led to a greater improvement .
Ordering type Mean
web 0.57
personalization 0.65
groupization 0.67
Smart Splitting Collaborative search tools should support division of
labor.In this we do following steps-
One member of the group enters a query term, which is then sent to a search engine.
The top results for this query on the basis of personalized score of every user are then divided up round-robin style amongst all of the group members.
Split searching can be used to allow group members to evaluate sets of results efficiently, without redundancy.
In this, we computed the normalized Discounted Cumulative Gain (DCG)(e.g.-for any query)
From this we can see that, DCG scores shows smart splitting performing better than the other two methods.
Split Method Mean (group queries)
Round Robin 0.59
Random 0.62
Smart 0.69
Group Hit-Highlighting Hit-highlighting is a technique used by most major search
engines to help users understand how relevant a result is to their information need.
Instances of the user’s keywords that appear within the title, snippet, or url of each search result in the results list are emphasized & all group members’ keywords that appear within a search result are also emphasized.
Then the rank is calculated by comparing the number of times those group queries’ keywords appeared within the result.
Conclusion The design of collaborative search systems can benefit
from reflecting on single-user techniques, such as personalization, and considering how they might be applied to groups. like now we are interested in studying explicit task based group.
Instantiating these concepts within a collaborative search tool is an important next step for understanding their impact on group dynamics and collaboration strategies
References Potential For Personalization: By Jaime Teevan, Susan T.
Dumais & Eric Horvitz, Microsoft Research.Understanding Groups’ Properties as a Means of
Improving Collaborative Search Systems: By Meredith Ringel Morris & Jaime Teevan ,Microsoft Research, Redmond, WA, USA.
Enhancing Collaborative Web Search with Personalization: Groupization,Smart Splitting,and Group Hit-Highlighting: By Meredith Ringel Morris, Jaime Teevan, &Steve Bush, Microsoft Research, Redmond, WA, USA
THANK YOU
QUERIES?