SoQL

1
SoQL A Language for Querying and Creating Data
in Social Networks

• Royi Ronen and Oded Shmueli
• Technion Israel Institute of Technology

March 29th, 2009 M3SN, Shenghai, China
2
Introduction

• As social networks become popular
• A lot of data
• Many participant
• Many connections
• Sizable participant record
• Proliferation to business and organizational
  cultures
• Many querying scenarios which can benefit from a
  domain-specific language
• SoQL is proposed as a step in this direction

3
Example
1
Charlie
Bob
5
2
4
Alice
3
Eve
Dave
9
6
7
Frank
Gloria
TF(id,weight)
8
0.7 1
0.6 2
0.4 3
0.4 4
0.3 5
0.9 6
0.85 7
0.85 8
0.5 9
TN(name,company,e-mail,position,experience)
4 Manager alice_at_hal.com HAL Alice
3 Manager bob_at_acme.net ACME Bob
2 Engineer cha_at_cia.gov CIA Charlie
5 Teacher dave_at_mtv.com MTV Dave
6 Scientist eve_at_acme.net ACME Eve
7 Technician fr_at_hal.com HAL Frank
6 Producer glor_at_abc.org ABC Gloria
4
Bobs Information Needs

• Bob works for ACME, and is looking for a job in
  HAL
• Bob is looking for a path which connects him to a
  manager in HAL, which in addition
• is at most 4 nodes long, and
• does not have any participant, except for Bob,
  working for ACME
• Results are to be ordered by the multiplication
  of weights along the path, excluding the first
  edge
• Higher quality social paths

5
Bobs query

• SELECT COUNT(PATH.nodes.), PATH
• FROM PATH (Bob TO X AS P1 TO Y AS P2)
• WHERE Y.company 'HAL' and
• Y.position 'manager' and
• ATMOST 0 IN P2.nodes SATISFY (company'ACME')
  and
• COUNT(P1.nodes.) 2 and
• COUNT(PATH.nodes.) lt 4
• ORDER BY MULT(P2.edges.weight)

The Path
Conditions on attributes
Path Predicates
Aggregation Path predicates
6
Result

• 4 (Bob, Dave, Gloria, Alice)
• 3 (Bob, Charlie, Alice)
• 4 (Bob, Dave, Eve, Alice)
• Multiplication values are 0.765, 0.3, 0.16

7
Model

• Undirected graph
• Reciprocal friends model
• Nodes and edges have attributes
• New Data Types
• Path An ordered set of distinct nodes, every
  two successive nodes are connected
• Group A set of nodes

8
Model

• Results are finite
• Social networks are constantly growing
• But finite at any point

9
Aggregation over Path/Group

• Aggregation over path/group is possible
• E.g., the number of nodes in path P1
• SELECT COUNT() FROM P1.nodes
• Or, as in the previous example
• MULT(P2.edges.weight)

10
Path Predicates

• ALL SATISFY (condition)
• ATMOST n
• ATLEAST n
• ALL EXCEPT UPTO n
• MAJORITY

11
Another information need

• Bob would like to find a group such that
• The group contains Bob and three others
• There exists a path of up to three edges from Bob
  to each of the three
• There exists a path of up to two edges between
  every two of the three
• All three have experience gt 5

12
SELECT FROM GROUP
Group with Paths
SELECT GROUP FROM GROUP (Bob AS G1,
DISTINCT(X,Y,Z) AS G2) WITH PATH (Bob TO X AS
P1), PATH (Bob TO Y AS P2), PATH (Bob TO Z AS
P3) WHERE COUNT(P1.edges.)lt3
and COUNT(P2.edges.)lt3 and COUNT(P3.edges.)lt
3 and ALL IN G2.nodes SATISFY (experiencegt5)
and ALL SUBGROUPS(U,V) IN G2 SATISFY (PATH(U
TO V AS P4) COUNT(P4.edges.)lt2))
IN
Aggregation on paths
Group Predicate
Subgroups
and COUNT(GROUP.nodes.)lt5
13
Group Predicates

• Group predicates refer to either
• nodes in a group or
• paths involving members of the group
• When referring to nodes, operators are the same
  as for paths
• ALL IN G2.nodes SATISFY (experiencegt5)
• When referring to paths, as in
• ALL SUBGROUPS(U,V) IN G2
  SATISFY(PATH(U TO V AS P4)
• COUNT(P4.edges.)lt2)
• operators are
• ALL SUBGROUPS,
• ATLEAST n SUBGROUPS,
• ALL EXCEPT UPTO n SUBGROUPS,
• MAJORITY SUBGROUPS

14
CONNECT

• Let R be a one-column relation of paths
• The paths are used for an automated process of
  referral intended to create a connection to the
  last node in the path

CONNECT USING PATH FROM R WHERE TIMEOUT36,
ATTEMPTS5, PARALLEL2, HISTORYtrue
15
CONNECT

• Let R be a one-column relation of groups
• An automated process will attempt
• Form a group, like, e.g., Facebook, or
• Create an edge between each pair in the group

CONNECT GROUP FROM R WHERE TIMEOUT48,
ATTEMPTS1, PARALLEL1
16
Implementation Issues

• Path/Group sizes are not necessarily predefined
  or known a priori
• Deployment parameters needed
• Maximum tuples in a result (Googles 1k)
• Maximal length of any path
• Maximal size of any group
• Time Limit

17
Finding paths

• Top-k self joins can be used to avoid large
  intermediate results
• In, e.g., distributed data, random walks can be
  used to extract candidates for paths in the
  result
• At any point, if the path can not satisfy the
  query, the walk aborts
• Many walking agents can provide a good
  approximation

18
Conclusions

• SoQL is a domain-specific, SQL-like query
  language for the social networks domain
• Creation of data is possible using the Path and
  the Group data types is possible
• Future work
• More expressive predicates, e.g., disjointness of
  two paths
• Implementation
• Advanced, optimized evaluation techniques for
  centralized and distributed environments

19
Thank You