Networks: Fall 2015

David Easley & Jon Kleinberg

Homework 1
Due 11:15am, Friday, September 11, 2015

As noted on the course home page, homework solutions must be submitted by upload to the
CMS site, at https://cms.csuglab.cornell.edu/. The file you upload must be in PDF
format. It is fine to write the homework in another format such as Word, as long as its
saved out as PDF. (From Word, for example, you can save files into PDF format.)
The CMS site will stop accepting homework uploads after the posted due date. We cannot
accept late homework except for University-approved excuses (which include illness, a family
emergency, or travel as part of a University sports team or other University activity).
Reading: The questions below are primarily based on the material in Chapters 2, 3, and 5
of the book. (Since Questions 4 and 5 involve terminology from Chapter 5, the background
for these will come from Wednesdays class.)

(1) Suppose that youre working with a group of sociologists who are studying the friendships among students in a small school. Students attend this school from grades 6 through
12, inclusive, and lets assume that there are 10 students in each grade. The sociologists
want to investigate the social network whose nodes are the students in the school (with 70
nodes, since there are 10 students in each of the 7 grades at the school).
Lets suppose that each pair of students whose grades differ by at most two is connected
by a strong tie, and and each pair of students whose grade differs by exactly three is connected
by a weak tie. (Note that under this rule, two students in the same grade are connected
by a strong tie, since their grades differ by zero.) There is no edge in the social network
connecting any two students whose grades differ by more than 3.
Suppose this is the information you know about the social network. Based on this, what
can you say about the Strong Triadic Closure Property in this network? Choose from among
the following possible answers:
(i) All nodes satisfy the Strong Triadic Closure Property. If this is your answer, explain
why all nodes satisfy the property.
(ii) Some nodes satisfy the Strong Triadic Closure Property and others do not. If this
is your answer, name a node in the network that satisfies the Strong Triadic Closure
Property, and name another node that violates the Strong Triadic Closure Property,
providing an explanation in both cases. (Since the nodes dont have specific names in
our description of the network, it is enough to describe the nodes you have in mind.)
(iii) No node in the network satisfies the Strong Triadic Closure Property. If this is your
answer, explain why there is no node that satisfies the property in other words, why
every node violates the Strong Triadic Closure Property.
1

(iv) Not enough information to tell. If this is your answer, explain why there isnt enough
information provided about the network to determine which of (i), (ii), or (iii) is the
case.

(2) One useful application of networks is to record how things overlap in time. Heres
an example of how this works.
Suppose a group of college students worked for different periods of time last summer as
interns at a small company. Figure 1 shows the dates when they were there.
Name of student
Rui
Sunita
Tal
Ursula
Victor

Dates at company
May 20 June 10
May 20 August 15
May 30 August 5
May 10 August 25
July 25 August 15
Figure 1:

Now, lets build a social network on these students, connecting two of them with an edge
if their dates at the company overlapped (that is, if there was at least one day when they
were both there). If they didnt overlap at all, then we dont connect them with an edge.
Furthermore, lets say that for two students connected by an edge, the edge between two
students is a strong tie if they overlapped at the company for at least a 15-day period, and
otherwise (if they overlapped, but for less than 15 days) the edge is a weak tie.
(a) Using this way of constructing the social network, draw the network on the five
students, labeling each edge as a strong or weak tie.
(b) Which students in the resulting network satisfy the Strong Triadic Closure Property,
and which do not? (Recall that unless a node violates the property, following the definition
from the book, it is said to satisfy it.) Provide an explanation for your answer.

(3) In the previous question, we took a time-line with a set of time intervals (the period
when each student was at the company) and built a graph from it: there was a node for each
time interval, and an edge between each pair of time intervals that overlapped.
Now lets consider going in the opposite direction: taking a graph, and asking whether
it corresponds to the overlaps among a set of time intervals.
Heres how a question like this might come up. Suppose youre working with a set of
historians, and theyre studying a set of documents from the sixteenth century that describe
a set of meetings of a council populated by members of the nobility that met over a few years.
2

A and B attended meetings

together
B and C attended meetings
together
C and D attended meetings
together

Figure 2:

Time

(a)

(b)

Figure 3:

Different members of the nobility belonged to the council for different intervals of time. The
documents dont have dates on them (or at least, the historians havent been able to figure
out the scheme for recording dates), but the documents do contain information about who
attended meetings together.
Heres what theyd like to do. For each person X in the documents, they plan to hypothesize a contiguous time interval during which X was a member of the council. (That
is, their hypothesis is that each person X had a start date when they joined the council,
and then they served continuously until their end date; no one ever left the council and then
joined it again later.) Theyd like to create these intervals in such a way that for every pair of
council members who are mentioned in the documents as attending a meeting together, their
time intervals overlap, and for every pair of council who are never mentioned as attending a
meeting together, their time intervals should not overlap.
So starting from the documents, they first create a graph: there is a node for each person
X, and for two people X and Y , there is an edge precisely when X and Y are mentioned as

having attended a meeting together. They then ask you whether you can create a set of time
intervals, one for each node in their graph, so that the time intervals for two nodes overlap
when they are connected by an edge, and the time intervals for two nodes do not overlap
when they are not connected by an edge. (Note that the historians are not concerned with
the distinction between strong ties and weak ties in their work, only with the presence or
absence of an edge.)
Heres an example for how this works. The historians start with a set of documents as
shown in Figure 2. They turn this into a graph as in Figure 3(a) showing that A and B
attended meetings together, B and C attended meetings together, and C and D attended
meetings together. But there is no evidence in the documents that A attended meetings
with C or D, nor is there evidence that B attended meetings with D, so there are no edges
between these pairs of nodes. Finally, they would like a set of time intervals as in Figure 3(b)
so that A overlaps B, B overlaps C, and C overlaps D, but A does not overlap C or D, and
B does not overlap D. That corresponds to a valid hypothesis for peoples membership on
the council, given the documents they have.
Now lets try this on two more complicated datasets.
(a) Suppose that based on the documents they have (about a different council with
different people than in the example above, but with the same premise), the historians
construct the graph in Figure 4

E
C

Figure 4:

Can you construct a set of time intervals, one for each node, so that the time intervals
overlap for each pair of nodes connected by an edge, and the time intervals do not overlap
for each pair of nodes not connected by an edge?
Give one of two possible answers:
(i) A set of time intervals with this property; or
(ii) A brief explanation for why it is not possible to construct a set of time intervals with
this property that is, why there cannot be a set of time intervals with the desired
pattern of overlaps corresponding to the edges in the graph. (In option (ii), you do
not need to provide a complete proof for why it is not possible; a brief explanation will
suffice for purposes of this question.)
4

A
D

B
C

Figure 5:

(b) Now consider a different set of documents, leading instead to the graph in Figure 5
Can you construct a set of time intervals, one for each node, so that the time intervals
overlap for each pair of nodes connected by an edge, and the time intervals do not overlap
for each pair of nodes not connected by an edge?
Give one of two possible answers:
(i) A set of time intervals with this property; or
(ii) A brief explanation for why it is not possible to construct a set of time intervals with
this property that is, why there cannot be a set of time intervals with the desired
pattern of overlaps corresponding to the edges in the graph. (In option (ii), you do
not need to provide a complete proof for why it is not possible; a brief explanation will
suffice for purposes of this question.)

+
-

+
-

Figure 6:

(4) (a) Consider the social network depicted in Figure 6, with each edge labeled as either
a friend or enemy relation. (The + denotes a friend relation and the - denotes an enemy
relation.) Is this network structurally balanced or not?
5

If you think it is balanced, describe how you could divide the nodes into two sets X and
Y , as in lecture and the book, so everyone in X is friends with each other, everyone in Y is
friends with each other, and everyone in X is an enemy of everyone in Y . If you think it is
not balanced, name a triangle in the network that does not satisfy the property of structural
balance.
-

+
-

+
-

Figure 7:

(b) Consider the social network depicted in Figure 7, with each edge labeled as either a
friend or enemy relation. (The + denotes a friend relation and the - denotes an enemy
relation.) Is this network structurally balanced or not?
If you think it is balanced, describe how you could divide the nodes into two sets X and
Y , as in lecture and the book, so everyone in X is friends with each other, everyone in Y is
friends with each other, and everyone in X is an enemy of everyone in Y . If you think it is
not balanced, name a triangle in the network that does not satisfy the property of structural
balance.

(5) Together with some anthropologists, youre studying a set of small villages in a
square-shape region of rain forest. Theyve imposed a two-dimensional coordinate system
on the region of rain forest: the anthropologists are based in a village that they label with
the point (0, 0), and then (x, y) in their coordinate system (with x 0 and y 0) refers to
a point that is x miles east and y miles north of their special designated village.
Suppose for simplicity in this story, there is a village each point of the form (x, y) where
x and y are each whole numbers between 0 and 5 inclusive.
Now, suppose that two villages view each other as allies if they are at most two miles
apart, and they view each other as enemies otherwise.
To study these relationships, the anthropologists build a signed complete graph on the
set of villages, with a + edge between each pair of villages that are allies and a - edge
between each pair of villages that are enemies. Does this graph satisfy the Structural Balance
6

property? Provide an explanation for your answer.

X
W

Z
T
P
S
V

Figure 8:

(6) Suppose youre consulting for a large social-networking company, helping them design
the part of their system that recommends new friendships to their members (i.e. to the users
of the system). The idea is to look at the position of a member Z in the overall social network,
and try to automatically recommend based on the pattern of links the name of one other user
to whom Z is not currently connected, but who Z might want to connect to. If Y is chosen
by the system as the recommendation for Z, then Z receives a prompt on his/her screen
asking if Z would like to add Y as a friend. The recommendation is viewed as successful if
Z accepts the suggestion and adds Y .
Youve been attending meetings in which the team in charge of the recommendation
system shows the results of the system on various real-life examples, and then you suggest
whether you think the system made the right recommendation. In this way, the recommendation team can try identifying deficiencies in their methods.
(a) In one example, the team shows you Figure 8, consisting of the user Z and all nodes
in the system who are at a distance of either 1 or 2 from Z in the social network. In this
example, the system recommended user S to Z. Do you think this was the best choice? If
so, explain why; if not, say what you think would be a better choice, and again explain why.
In either case, relate your explanation to principles from class.
(b) Currently the recommendation system is based on looking at small portions of the
social network like the one in Figure 8. But the recommendation team also has a heuristic
that estimates, for each edge between two people, how strong a tie it is. (The estimate is
7

based on measures of the amount of activity that goes on between the two people.) Theyre
currently not using these estimates of tie strength, but they are interested in whether theres
a way to incorporate them into the recommendation system to improve the quality of the
recommendations.
Suggest how the recommendation system might make use of information about how
strong each tie is, and why it might help the recommendations compared to having just a
pure unlabeled network in which all edges are treated equally. In your explanation, you
should refer back to principles from class and/or the book.