1 DEFINITIONSMATIVE ANSWERS THAN A SINGLE PHRASE. IN FACT, IT ISIN T...
2.1 Definitions
mative answers than a single phrase. In fact, it is
In this subsection we define some concepts that
often the case, that a complex multi-sentence ques-
would be helpful to clarify our problems. First we
tion could be answered from multiple aspects by dif-
define a complex multi-sentence question as a ques-
ferent people focusing on different sub questions.
tion with the following properties:
Therefore we address the incomplete answer prob-
lem by developing a novel summarization technique
Definition: A complex multi-sentence question
is one that contains multiple sub-questions.
taking different sub questions and contexts into con-
In the cQAs scenario a question often consists of
sideration. Specifically we want to learn a concise
one or more main question sentences accompany by
summary from a set of corresponding answers as
some context sentences described by askers. We
supplement or replacement to the “best answer”.
treat the original question and context as a whole
We tackle the answer summary task as a sequen-
single complex multi-sentence question and obtain
tial labeling process under the general Conditional
the sub questions by question segmentation. We
Random Fields (CRF) framework: every answer
then define the incomplete answer problem as:
sentence in the question thread is labeled as a sum-
Definition: The incomplete answer problem is
mary sentence or non-summary sentence, and we
one where the best answer of a complex multi-
concatenate the sentences with summary label to
sentence question is voted to be below certain star
form the final summarized answer. The contribution
ratings or the average similarity between the best an-
of this paper is two-fold:
First, we present a general CRF based framework
swer and all the sub questions is below some thresh-
olds.
predicting asker-rated quality of answers was evalu-
ated by using a logistic regression model. However,
We study the issues of similarity threshold and the
to the best of our knowledge, there is no work in
minimal number of stars empirically in the experi-
mental section and show that they are useful in iden-
evaluating the quality of features for community an-
swer summarization. In our work we model the fea-
tifying questions with the incomplete answer prob-
lem.
ture learning and evaluation problem as a group L
1