Database Reference
In-Depth Information
l
SIMILAR
asks the system to find users similar to us, and share the data with
them only. Similarity between two users
u
and
u
0
can be defined in several ways:
based on commonality of queries (that is, common data of interest) or even
commonality of private data (which the system has access to). This includes
similarity not only of data (using references) but also of tags, annotations,
external links. While we do not discuss this in detail, we point out that capturing
private data allows a more exact determination of the similarity among users.
Formally, given user
u
, the
history
of
u
is a sequence
H
u
¼
,(
a
n
,
t
n
))
of pairs, where the
a
i
are
database accesses
and
t
i
are timestamps. Each database
access
a
i
is either a read access or query
r
i
or a write access (insert, delete,
update)
w
i
. The history captures the interaction of
u
with the database by storing
the commands that
u
issued against the database and the time at which each
command was issued. A
user profile
for
u
is defined as a pair
Pf
u
¼
((
a
1
,
t
1
),
...
(
u
))
where
H
u
is the history of
u
and
P
u
is the private view of
u
. Similarity between
users
u
and
u
0
could be established based solely on
H
u
and
H
0
; however, it seems
clear that a similarity that takes into account
Pf
u
and
Pf
u
0
is likely to be more fine
grained and return better results.
(
H
u
,
P
v
When the user decides to share private data, the private view is copied in a system-
wide table which has a similar schema, but with an attribute
user id
added, so that
the author of the content can be identified. A command
SUBSCRIBE
should also be
supported by the system, so that users can gain access to this system-wide table. Note
that this command can be extended similarly to the share command we described so
that users are able to request access to everyone's private data, or just some specific
users. Note also that since the sharing is done through an intermediate system-wide
table, the users have control about data they access and which data they share. This is
an extremely desirable property; giving users control over their data they generate
encourages them to create such data in the first place. Note also that the system may
impose some constraints of its own. The most important (and obvious) one is that
content may be shared only with users who are allowed to see the referents, that is,
that have (at least, read) access to the data being referred to. Allowing the users to see
metadata on data that they are not allowed to access may be considered a security
breach especially since some user-created content may give away the data. Thus,
permissions given by the user may be narrowed down by the system to that part of the
audience who has permission to access the underlying data. Note that, in computing
similarity between pairs of users, most measures along the lines sketched here would
return a very low similarity between users who cannot access at least some common
data, and hence the risk of sharing data with unauthorized users is quite small.
7.5.4 Private View Maintenance
Finally, we discuss the issue of maintaining the private view over time. We focus on
changes to referred data or metadata.
Search WWH ::
Custom Search