IMPIMP9

Reviewer Problem findings:

1. Matching algorithm ( the main problem)

2. Limits of number of proposals and the time available with each reviewer (to be taken care
while making the alogorithm)
3. The reviewer group has subsets(groups within the broader group)or not impacts the review
algorithm and process
4.
5.

dirk thorleuchter idea mining

Topic models, reviewer finding, expert retrieval ,

We follow previous approaches in treating expert finding as an information retrieval task

You can break a recommender down into a classification ore regression problem

I would call recommender a higher-order system that internally is solving regression or

classification problems. Do you agree?

Group decision making ?

what is word or text stemming ?

maximum-weighted matching algorithm of Kuhn and Munkres (also known as the Hungarian
Algorithm)
what is regression ?

what is heuristics?

Ontology-driven topic inference using automatically- extracted keywords

(VSM) technique and machine learning

The same idea can be extended to represent the profile of a reviewer as long as we can crawl and
capture a set of publications written by a person/reviewer from common sources such as, WWW,
Google Scholar, ACM Digital Library, Citeseer and other repositories.

How to calculate term weights for words in VSM model?

Can we use NLP and machine learning to extract keywords and match (ca use existing algorithm,
just creates keywords dynamically)

domain-ontology based topic-inference to represent papers and reviewers more precisely over a
vector space comprising of topic-vectors,

Keyphrase Extraction Tool, KEA(uses machine learning)

https://www.textrazor.com(keyword
maker)

http://www.cisp-
bmei.cn/Submission/openconf.php(review
system)
IMP6
Author-Topic model :
3.2.1 Author-Topic model For this paper,
we evaluate two topic models. The first is
a simplified version of the AT model. All
training documents in the corpus are
constrained to have a single author, so the
variables representing which author is
responsible for a given word are
meaningless. The resulting model can be
thought of as a simple topic model run on
the concatenated documents described
earlier in the language model section for
the single-document author model. The
generative model for the single-author AT
model can be described by the following
Bayesian hierarchical model. The model
includes two Dirichlet hyperparameters, α
and β, which are the size of the set of
topics and the vocabulary of the corpus,
respectively

Author-Persona-Topic model
In addition to the single-author AT model,
we present a novel topic model, the
Author-Persona-Topic (APT) model. The
difference between APT and AT is that
rather than grouping all papers by a given
author under a single topic distribution, we
allow each author’s documents to be
divided into one or more clusters, each
with its own separate topic distribution.
These clusters represent different
“personas”under which a single author
writes, each
Rexa database of research paper