List of projects: Talk to me in order to get more details about these projects.
| # | Project | May Lead to
MSc | Taken by | Brief Description |
|---|
| 1 |
Conference Ranking |
Yes |
Yavar
+
Wei
+
Camilo |
| Conferences are currently ranked subjectively by
reputation. The project consists of devising a metric to rank
conferences based on their paper citation, the reputation of the
authors publishing in them, etc. This information could be extracted
from scholar.google or citeseer in addition to the public database
dblp.
NB. There are many possibilities for
ranking. We can either have different projects on this topic or a
team project testing different approaches.
|
|---|
| 2 |
Asymmetric Parallel Frequent Itemset Mining |
Yes |
Shuang |
Typically, a parallel data mining program runs the same
program on all processors. This project consists of identifying
transactional data features that would identify the most appropriate
algorithm to run on each processor given its data partition. |
|---|
| 3 |
Emerging Sequences |
Yes |
Kang |
Comparing sets of sequences is relevant to many
applications. This project consists of implementing algorithms to
identify contrasting sequences among sets of sequences. |
|---|
| 4 |
Contrast Sets based on Association Rules |
yes |
Yi |
Contrast sets can be mined by some specific algorithms such as
STUCCO and CIGAR. They can also be mined by means of association rule
mining. The project is to use association rule mining to extract
contrast sets and see whether contrast sets can be used to
characterize clusters or improve classifiers. |
|---|
| 5 |
Home Page Detection |
Possible |
Fariba |
This project consists of identifying a personal web page of a
given person if this web page exists. This can be done by using a
search engine to extract pages using a person's name as query and
classify the resulting pages into Home page and non-home page. |
|---|
| 6 |
Identity resolution by authorship promiscuity |
Yes |
Zhiyu |
In a social network of paper authors the same names can
represent different authors, but can we resolve this identification
issue by
means of authorship promiscuity. Authors are "promiscuous" if they
co-author with different groups of authors for each context
(i.e. field or group of conferences). |
|---|
| 7 |
Query Refinement Using Social Network Analysis |
Yes |
Ying |
The project consists of exploiting community mining in social
network analysis to cluster and label documents in search engine
results in order to provide suggested refinement to search queries. |
|---|
| 8 |
Associative classifier by row-enumeration for high dimensionality |
Yes |
Jiaofen |
Associative classifiers use association rule mining to
discover classification rules. However, for a high dimensional space,
association rule mining can be too expensive to be
practical. Row-enumeration is a new alternative. Can it be used to
build a classifier for data such as microarray data with many features
but few patients? |
|---|
| 9 |
Mammography Classification |
Yes |
Farzaneh |
An associative classifier should be used to classify
mammograms into cancerous and normal cases. An interface is to be
built receiving mammograms and simulating the diagnostic is to be built. |
|---|
| 10 |
Feature Space Selection |
Yes |
Naimeh
+
Mojdeh |
When classifying records with a large number of attributes
(high dimensionality) it is necessary to select the most relevant ones
to build the classification model. There are many techniques for
dimensionality reduction, feature selection, etc. This project is to
explore two new hypothesis for feature selection: one using subspace
clustering and the other one using random walks in bipartite graphs.
NB. There could be two individual projects
or a team project testing these different hypothesis.
|
|---|
| 11 |
Question-Answer recommender System |
Yes |
|
In the context of a search engine, one could send a question
instead of keywords. The idea is to expend the question, retrieve
pertinent documents using search engines, then summarize these
resources and provide the concise summaries as recommended answers to
the original question. |
|---|
| 12 |
Classification Explanation Generation for Associative Classifiers |
Possible |
Dave |
When a classifier predicts a class for a new object a straight
forward question is "how was the conclusion drawn?" The project is to
built an UI that would explain the results of an associative classifier. |
|---|