Choosing the right crowd. Expert finding in social networks. edbt 2013
1. CHOOSING
THE RIGHT CROWD
EXPERT FINDING IN SOCIAL NETWORKS
Alessandro Bozzon
Marco Brambilla
Stefano Ceri
Matteo Silvestri
Giuliano Vesci
Politecnico di Milano
Dipartimento di Elettronica, Informazione e BioIngegneria
2. Problems and terms
• Human Computation:
• Computation carried out by groups of humans (examples: collaborative
filtering, online auctions, tagging, games with a purpose)
• Crowd-sourcing:
• The process of building a human computation using computers as
organizers, by organizing the computation as several tasks (possibly with
dependences) performed by humans
• Crowd-searching:
• A specific task consisting of searching information
• Crowd-sourcing Platform:
• A software system for managing tasks, capable of organizing tasks,
assigning them to humans, assembling and processing returned results
(such as Amazon Mechanical Turk, Doodle)
• Social Platform:
• A platform where humans perform social interactions (such as Facebook,
Twitter, LinkedIn)
4. Why Crowd-search?
• People do not trust web search completely
• Want to get direct feedback from people
• Expect recommendations, insights, opinions, reassurance
5. And given that crowds spend times on
social networks….
• Our proposal is to use social networks and Q&A websites as
crowd-searching platforms, in addition to crowdsourcing
platforms
• Example: search tasks
Query Answer
Query Interface
Search Execution Social
Human Interaction
Engine Networks
Management
Q&A
Local
Human
SE Access Source Crowd-
Access
Interface Access source
Interface
Interface platforms
6. From social workers to communities
• Issues and problems
• Motivation of the responders
• Intensity of social activity of the asker
• Topic appropriateness
• Timing of the post (hour of the day, day of the
week)
• Context and language barrier
7. Crowd-searching after conventional
search
• From search results to friends and experts feedback
initial query
Human
Search
System Search
System
Social Platform
Social Platform Social Platform
14. Task management problems
Typical crowdsourcing problems:
• Task splitting: the input data collection is too complex relative
to the cognitive capabilities of users.
• Task structuring: the query is too complex or too critical to be
executed in one shot.
• Task routing: a query can be distributed according to the
values of some attribute of the collection.
Plus:
• Platform/community assignment: a task can be assigned to
different communities or social platforms based on its focus
15. Task Design
• Which are the input objects of the crowd interaction?
• Should they have a schema (set of fields, each defined by a name and a type)?
• Which operations should the crowd perform?
• Like, label, comment, add new instances, verify/modify data, order, etc.
• How should the task be split into micro-tasks assigned to each person?
How should a specific object be assigned to each person?
• How should the results of the micro-tasks be aggregated?
• Sum, Average, Majority voting, etc.
• Which execution interface should be used?
16. Operations
• In a Task, performers are required to execute logical operations on input objects
• e.g. Locate the faces of the people appearing in the following 5 images
• CrowdSearcher offers pre-defined operation types:
• Like: Ask a performer to express a preference (true/false)
• e.g. Do you like this picture?
• Comment: Ask a performer to write a description / summary / evaluation
• e.g. Can you summarize the following text using your own words?
• Tag: Ask a performer to annotate an object with a set of tags
• e.g. How would you label the following image?
• Classify: Ask a performer to classify an object within a closed-set of alternatives
• e.g. Would you classify this tweet as pro-right, pro-left, or neutral?
• Add: Ask a performer to add a new object conforming to the specified schema
• e.g. Can you list the name and address of good restaurants nearby Politecnico di Milano?
• Modify: Ask a performer to verify/modify the content of one or more input object
• e.g. Is this wine from Cinque Terre? If not, where does it come from?
• Order: Ask a performer to order the input objects
• e.g. Order the following books according to your taste
17. Splitting Strategy
• Given N objects in the task
• Which objects should appear in each MicroTask?
• How many objects in each MicroTask?
• How often an object should appear in MicroTasks?
• Which objects cannot appear together?
• Should objects be presented always in the same order?
18. Assignment Strategy
• Given a set of MicroTasks, which performers are assigned to them?
• Online assignment
• Micro Tasks dynamically assigned to performers
• First come / First served
• Based on a choice of the performer
• Offline assignment
• MicroTasks statically assigned to performers
• Based on performers’ priority
• Based on matching
• Invitation
• Send an email to a mailing list
• Publish a HIT on Mechanical Turk (dynamic)
• Create a new challenge in your game
• Publish a post/tweet on your social network profile
• Publish a post/tweet on your friends' profile
19. Deployment: search on the social network
• Multi-platform deployment
Generated query template
Embedded External
Native Standalone
application application application
API
Social/ Crowd platform
Native
Embedding behaviours
Community / Crowd
24. Crowdsearch experiments
• Some 150 users
• Two classes of experiments:
• Random questions on fixed topics: interests (e.g. restaurants in the vicinity of
Politecnico), to famous 2011 songs, or to top-quality EU soccer teams
• Questions independently submitted by the users
• Different invitation strategies:
• Random invitation
• Explicit selection of responders by the asker
• Outcome
• 175 like and insert queries
• 1536 invitations to friends
• 230 answers
• 95 questions (~55%) got at least one answer
31. Problem
• Ranking the members of a social group according
to the level of knowledge that they have about a
given topic
• Application: crowd selection (for Crowd Searching
or Sourcing)
• Available data
• User profile
• behavioral trace that users leave behind them through
their social activities
32. Considered Features
• User Profiles
• Plus Linked Web Pages
• Social Relationships
• Facebook Friendship
• Twitter mutual following relationship
• LinkedIn Connections
• Resource Containers
• Groups, Facebook Pages
• Linked Pages
• Users who are followed by a given user are resource containers
• Resources
• Material published in resource containers
38. Resource Processing
URL Entity
Content Recognition and
Extraction Disambiguation
Resource Language
Extraction Identification
Text
Processing
• Extraction from Social • Text Processing
Network APIs
• Sanitization, tokenization,
stopword, lemmatization
• Extraction of Text from linked
Web Pages
• Alchemy Text Extraction APIs • Entity Extraction and
Disambiguation
• Language Identification
• TagMe
39. Method – Resource Score
Entity Component
Weighting
• tf(t,r) term frequency -- irf(t) inverse resource frequency of t
• ef(e,r) entity frequency -- eir(e) inverse entity frequency of e
we(e,r) relevance of entity in resource
40. Method: Expert Score
Resource weight for
Window Size given expertise
Resource Score
• Experts are ranked according to score(q,ex)
41. Dataset
• 7 kinds of expertises
• Computer Engineering, Location, Movies & TV, Music, Science,
Sport, Technology & Videogames
• 40 volunteer users (on Facebook & Twitter & LinkedIN)
• 330.000 resources (70% with URL to external resources)
• Groundtruth created trough self-assessment
• For expertise need, vote on 7 Likert Scale
• EXPERTS expertise above average
42. Distribution of Expertise and Resources
• Avg Expertise ~ 3.5 / 7
• High Music and Sport
Expertise
• Low Location Expertise
30
Experts Expertise
7.0
# Dom Experts Avg Dom Expertise
25 Avg # Experts Avg Expertise 6.5
6.0
Avg. Expertise
20 5.5
# Experts 15
5.0
4.5
10 4.0
3.5
• High # Resources on Facebook 5 3.0
and Twiitter 2.5
0 Co L M S M S T
mp oca ovie cien usic port echn
• Higher # users on Facebook ute tion
r
ce o lo
gy
Domains
43. Metrics
• We obtain lists of candidate experts and assess them
against the ground truth, using:
• For precision:
• Mean Average Precision (MAP)
• 11-Point Interpolated Average Precision (11-P)
• For ranking:
• Mean Reciprocal Rank (MRR) – for the first value
• Normalized Discounted Cumulative Gain (DCG) – for more values, can
be set @N for the first N values
46. Social Network Analysis
•a
Comparison of the results obtained with All the social networks, or separately by
FaceBook, TWitter, and LinkedIn.
47. Main Results
• Profiles are less effective than level-1 resources
• Resources produced by others help in describing each individual’s
expertise
• Twitter is the most effective social network for expertise
matching – sometimes it outperforms the other social
networks
• Twitter most effective in Computer Engineering, Science, Technology &
Games, Sport
• Facebook effective in Locations, Sport, Movies & TV, Music
• Linked-in never very helpful in locating expertise
49. Main Message
• Crowd-sourcing should be dynamically adapted
• The best way to do so is through active rules
• Four kinds of rules:
execution / object / performer / task control
• Guaranteed termination
EXECUTION
• Extensibility
OBJECT PERFORMER TASK
CONTROL CONTROL CONTROL
OBJECT PERFORMER TASK
Control Production Rules
Result Production Rules
Execution Modifier Rules
Editor's Notes
LOCAL SOURCE: sorgentidatilocalisfruttate dal Search Execution Engine, magariaccedutedallo Human Interaction Management per configurare / gestirei task. La suaesistenza e’ accessoriarispettoaglialtri, e codificainformazioni applicative specificheICONE DI DX, DALL’ALTO a SX(social networks) Facebook, Twitter, Google +(Q&A systems) StackOverlflow, Yahoo Answers, Quora(HC Platforms) Freebase, Amazon Mechanical Turk, ODesk
MANUALE + RANDOM = ALL Non arriva a 100 perche’ sulle X ci siamofermati ad 1 giornodalladomanda (alcunerispostepotrebberoesserearrivatedopo)
LIKE + ADD = ALL Non arriva a 100 perche’ sulle X ci siamofermati ad 1 giornodalladomanda (alcunerispostepotrebberoesserearrivatedopo)
OWNERSHIP siapplica a risorse create in spazi DI PROPRIETA’ dell’utente, come ilsuo MURO Facebook, I GRUPPI da luicreati, la SUA Timeline Twitter, etc. CREATE invecesiapplica a risorse create suspazialtrui. Per esempio, il MURO di altri
Mean and Average are near mildly skewed distribution