--Presented by Tianyi Zhang Building Community Wikipedias: A Machine-Human Partnership Approach

Preview:

Citation preview

--Presented by Tianyi Zhang

Building Community Wikipedias: A Machine-Human Partnership

Approach

MotivationWikipedia portal involves deeper and

broader in our lives.Existing Approaches

Semi-Automatic Methods(such as Libra, Citeseer) to extract and integrate data from a multitude of data sources.

Deploy an initial portal in wiki format, then invite community members to revise and add material.(such as Wikipedia)

The Ultimate goal of this paper

Problems during the path to endHow to build an initial wiki portal?

How to model and implement the structure database G?

What should be the view language?How to export the structured data of the views into

wiki pages?How to manage user contributions to the portal?

What is it that U is conceptually allowed to edit?How to efficiently infer such edits based on what U

has done to a wiki page W?User can edit schema, not just the data.User’s edit limited to this wiki interface is usually

ambiguous.

Contribution of this paperIntroduce a novel hybrid approach which

integrates user’s insight and machines input.Provide solution to modeling the underlying

structured database, and exporting these views in wiki pages.

Employ an efficient solution to process user edits from wiki page to underlying databases.

Empirical results over a real-world implementation to demonstrate the promise of the approach

Architecture

Problems during the path to endHow to build an initial wiki portal?

How to model and implement the structure database G?

How to model and implement the structure database G?

Has to be intuitive…The author define schema Gs of G to

consist of a set of entity types E1,…En and a set of relation types R1,…Rn

The author defines the data Gd of G to be a temporal ER data graph.A set of nodes that specify entity instancesA set of edges that specify relation instances

Example

Storing G using RDBMSCreate several tables to store

definitions of entity types, relationship types and their attributes (such as person has attributes name, title, and organization).

Create table Entity_ID(id, etype) and Relationship_ID(id, rtype, eid1, eid2)

A_M and A_U are two tables which are to keep track of all values from users and machines.

Human-first policy

Example

Problems during the path to endHow to build an initial wiki portal?

How to model and implement the structure database G?

What should be the view language?

View Language RequirementsEvery wiki page is to describe one entity or

relationshipMaterialize quicklyMortify view quickly

A path-based view languageData path and schema path

Data path-start from one entity to another entity, retains only certain attributes for each node/edge along the path

Schema path

Example

A path-based ER view

Example

Creating Views over ER GraphWikify all entity and relationshipSpecify a default view schema Ve that

specifies a neighborhood of instances of E

Problems during the path to endHow to build an initial wiki portal?

How to model and implement the structure database G?

What should be the view language?How to export the structured data of the

views into wiki pages?

Converting Views to Wiki PagesThe straightforward solution is to convert

Vd into a set of natural-language sentences. But, it is difficult to extract and update structured data.

Problems during the path to endHow to build an initial wiki portal?

How to model and implement the structure database G?

What should be the view language?How to export the structured data of the

views into wiki pages?How to manage user contributions to the

portal?What is it that U is conceptually allowed to

edit?

What is it that U is conceptually allowed to edit?

Problems during the path to endHow to build an initial wiki portal?

How to model and implement the structure database G?

What should be the view language?How to export the structured data of the views into

wiki pages?How to manage user contributions to the portal?

What is it that U is conceptually allowed to edit?How to efficiently infer such edits based on what U

has done to a wiki page W?User can edit schema, not just the data.User’s edit limited to this wiki interface is usually

ambiguous.

Infer & Execute Structured Edits

Example

Mapping User Edits into Sequence of Actions

Empirical Evaluation

Empirical Evaluation

ThanksQuestion

Recommended