Information Technology Reference
In-Depth Information
To model such effects, the daily cumulative numbers of “posts”, “shares”, “likes”,
“comments” on the official Facebook Fan Pages of the dramas are included as fea-
tures.
3. Opinion Polarity Features
Users may express their thoughts toward a show via a Facebook Fan Page, and such
opinions will have influence on others' opinions. For example, if most of the fans are
looking forward to the upcoming episode, it will be reasonable to assign a higher
audience rating for the new episode. Thus we propose to analyze the polarity (i.e.
positive or negative) of users' posts and replies on Facebook fans page. We use the
daily cumulative number of positive and negative words in both posts and comments
as the opinion polarity features.
4. Trend Features
Google Trends is a useful tool provided by Google Inc. to investigate the popularity
of a keyword in a region. Given certain time period, it gives the number of searches
for a keyword relative to the total number of searches across this period. The dis-
played number is normalized such that the highest number is equal to 100 and the
lowest number is equal to zero. For each drama, we collect time series data from
Google Trends for three different sets of keywords. We use drama name as well as
actor/actress's name as queries in Google Trends to obtain the corresponding features.
4
Methodology
4.1
Gaussian Process Regression (GPR)
A typical regression problem can be formulated as
(1)
where is the input vector, is the observed target value, and is a function that
models the underlying process of generating the data points , :1,…, .
An additive independent and identically distributed Gaussian noise ~ 0, is
assumed.
There are two equivalent ways to derive the predictive distribution for Gaussian
process regression, namely the weight-space view and the function-space view [13].
In the following paragraphs, we will give a brief introduction to the main concepts of
GPR in the function-space.
A random process :∈ is defined as a collection of random va-
riables indexed by an ordered set . In the audience rating prediction problem,
we consider the input space , where is the dimension of input vectors.
The random variable therefore represent the value of the random function
evaluated at the data point . If normality is assumed, the random process is called a
Gaussian process (GP). An important property of GP is that any finite collection of
the random variables will be jointly normally distributed.
Search WWH ::




Custom Search